Model results
OSWorld
Choose a shared evaluation protocol to compare model scores. Versions, sources and configurations stay separate.
Atlas snapshot captured 10/19/2026, 10:20:28 AM UTC. This is an import timestamp, not the evaluation date. Coverage may be incomplete.
Reviewed recommendation and limits
Start with OSWorld for long desktop workflows. Select the task release and compare binary completion separately from partial credit.
Publisher-declared end-to-end results for June 24, August 8 and v2.1 task releases; binary completion and partial credit; full or offline datasets; exact step and tool settings. Imported groups below may use different protocols; this review does not certify each run. Full and offline datasets, task releases, step budgets, effort and tool settings stay separate. Provider effort labels do not imply equal compute. Exact execution backends and measurement dates are unreported; Muse Spark task denominators remain unresolved.
Evaluation groups
Open a group to see its exact protocol and scores. Different configurations remain separate.
Model rankings
Open a group to compare scores within its reported protocol. Positions do not establish statistical significance.
- OSWorld binary completionv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning max; tool setting batch tool; binary co…
Task release v2026.08.08; full dataset; 500 steps; reasoning max; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld binary completionv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning high; tool setting batch tool; binar…
Task release v2026.08.08; offline dataset; 500 steps; reasoning high; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld binary completionv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning low; tool setting batch tool; binary…
Task release v2026.08.08; offline dataset; 500 steps; reasoning low; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld binary completionv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning max; tool setting batch tool; binary…
Task release v2026.08.08; offline dataset; 500 steps; reasoning max; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld binary completionv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning medium; tool setting batch tool; bin…
Task release v2026.08.08; offline dataset; 500 steps; reasoning medium; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld binary completionv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; bina…
Task release v2026.08.08; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld partial creditv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning max; tool setting batch tool; partial c…
Task release v2026.08.08; full dataset; 500 steps; reasoning max; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning high; tool setting batch tool; parti…
Task release v2026.08.08; offline dataset; 500 steps; reasoning high; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld partial creditv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning low; tool setting batch tool; partia…
Task release v2026.08.08; offline dataset; 500 steps; reasoning low; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld partial creditv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning max; tool setting batch tool; partia…
Task release v2026.08.08; offline dataset; 500 steps; reasoning max; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld partial creditv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning medium; tool setting batch tool; par…
Task release v2026.08.08; offline dataset; 500 steps; reasoning medium; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld partial creditv2026.08.08
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; part…
Task release v2026.08.08; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- OSWorld binary completionv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning enabled; tool setting standard; binary …
Task release v2026.06.24; full dataset; 150 steps; reasoning enabled; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld binary completionv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning max; tool setting standard; binary comp…
Task release v2026.06.24; full dataset; 150 steps; reasoning max; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld binary completionv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning enabled; tool setting standard; binary …
Task release v2026.06.24; full dataset; 300 steps; reasoning enabled; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld binary completionv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning max; tool setting standard; binary comp…
Task release v2026.06.24; full dataset; 300 steps; reasoning max; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld binary completionv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning enabled; tool setting standard; binary …
Task release v2026.06.24; full dataset; 500 steps; reasoning enabled; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld binary completionv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting batched tool; binary …
Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting batched tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld binary completionv2026.06.24
OSWorld (reviewed publisher results)
3 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting standard; binary comp…
Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning enabled; tool setting standard; partial…
Task release v2026.06.24; full dataset; 150 steps; reasoning enabled; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning max; tool setting standard; partial cre…
Task release v2026.06.24; full dataset; 150 steps; reasoning max; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning enabled; tool setting standard; partial…
Task release v2026.06.24; full dataset; 300 steps; reasoning enabled; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning max; tool setting standard; partial cre…
Task release v2026.06.24; full dataset; 300 steps; reasoning max; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning enabled; tool setting standard; partial…
Task release v2026.06.24; full dataset; 500 steps; reasoning enabled; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.06.24
OSWorld (reviewed publisher results)
2 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting batched tool; partial…
Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting batched tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- OSWorld partial creditv2026.06.24
OSWorld (reviewed publisher results)
3 models with scores
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting standard; partial cre…
Task release v2026.06.24; full dataset; 500 steps; reasoning max; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
Other reported results
These reports have no model ranking. Open a group for its evidence and the reason positions are withheld.
- Claude Sonnet 4verified
OSWorld · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date 2025-07-28
Scope · Verified; approach=General model; max steps=15; a11y tree=No; coding actions=No; multiple rollouts=N…
Verified; approach=General model; max steps=15; a11y tree=No; coding actions=No; multiple rollouts=No; success/total=112.49/361; paper=https://www.anthropic.com/news/claude-4
- Claude Sonnet 4verified
OSWorld · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date 2025-07-28
Scope · Verified; approach=General model; max steps=50; a11y tree=No; coding actions=No; multiple rollouts=N…
Verified; approach=General model; max steps=50; a11y tree=No; coding actions=No; multiple rollouts=No; success/total=158.42/361; paper=https://www.anthropic.com/news/claude-4
- Claude Sonnet 4verified
OSWorld · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date 2025-07-28
Scope · Verified; approach=General model; max steps=100; a11y tree=No; coding actions=No; multiple rollouts=…
Verified; approach=General model; max steps=100; a11y tree=No; coding actions=No; multiple rollouts=No; success/total=149.41/361; paper=https://www.anthropic.com/news/claude-4
- o3verified
OSWorld · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date 2025-07-28
Scope · Verified; approach=General model; max steps=15; a11y tree=No; coding actions=No; multiple rollouts=N…
Verified; approach=General model; max steps=15; a11y tree=No; coding actions=No; multiple rollouts=No; success/total=32.85/361; paper=https://openai.com/index/introducing-o3-and-o4-mini/
- o3verified
OSWorld · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date 2025-07-28
Scope · Verified; approach=General model; max steps=50; a11y tree=No; coding actions=No; multiple rollouts=N…
Verified; approach=General model; max steps=50; a11y tree=No; coding actions=No; multiple rollouts=No; success/total=61.99/361; paper=https://openai.com/index/introducing-o3-and-o4-mini/
- o3verified
OSWorld · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date 2025-07-28
Scope · Verified; approach=General model; max steps=100; a11y tree=No; coding actions=No; multiple rollouts=…
Verified; approach=General model; max steps=100; a11y tree=No; coding actions=No; multiple rollouts=No; success/total=83.03/361; paper=https://openai.com/index/introducing-o3-and-o4-mini/
- Claude Opus 4.5verified
OSWorld · OSWorld via Epoch AI
Single-model result · Positions withheld
Measurement date not reported
Scope · result date not supplied
- Claude Sonnet 4.5verified
OSWorld · OSWorld via Epoch AI
Single-model result · Positions withheld
Measurement date not reported
Scope · 50 steps · result date not supplied
- Kimi K2.5verified
OSWorld · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · 100 steps · result date not supplied
- OSWorldverified
OSWorld via Epoch AI
2 models with scores · Positions withheld
Measurement date not reported
Scope · 15 steps · result date not supplied
- OSWorldverified
OSWorld via Epoch AI
3 models with scores · Positions withheld
Measurement date not reported
Scope · 100 steps · result date not supplied
- Claude Opus 5v2026.08.08
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning high; tool setting batch tool; binary c…
Task release v2026.08.08; full dataset; 500 steps; reasoning high; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2026.08.08
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning low; tool setting batch tool; binary co…
Task release v2026.08.08; full dataset; 500 steps; reasoning low; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2026.08.08
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning medium; tool setting batch tool; binary…
Task release v2026.08.08; full dataset; 500 steps; reasoning medium; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2026.08.08
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary …
Task release v2026.08.08; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2026.08.08
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning high; tool setting batch tool; partial …
Task release v2026.08.08; full dataset; 500 steps; reasoning high; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2026.08.08
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning low; tool setting batch tool; partial c…
Task release v2026.08.08; full dataset; 500 steps; reasoning low; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2026.08.08
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning medium; tool setting batch tool; partia…
Task release v2026.08.08; full dataset; 500 steps; reasoning medium; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2026.08.08
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial…
Task release v2026.08.08; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- GPT-5.6 Solv2026.08.08
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning none; tool setting batch tool; binar…
Task release v2026.08.08; offline dataset; 500 steps; reasoning none; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- GPT-5.6 Solv2026.08.08
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; offline dataset; 500 steps; reasoning none; tool setting batch tool; parti…
Task release v2026.08.08; offline dataset; 500 steps; reasoning none; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Sonnet 4.6v2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning medium; tool setting standard; binary c…
Task release v2026.06.24; full dataset; 150 steps; reasoning medium; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Sonnet 4.6v2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning medium; tool setting standard; binary c…
Task release v2026.06.24; full dataset; 300 steps; reasoning medium; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Sonnet 4.6v2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning medium; tool setting standard; binary c…
Task release v2026.06.24; full dataset; 500 steps; reasoning medium; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Sonnet 4.6v2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning medium; tool setting standard; partial …
Task release v2026.06.24; full dataset; 150 steps; reasoning medium; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Sonnet 4.6v2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning medium; tool setting standard; partial …
Task release v2026.06.24; full dataset; 300 steps; reasoning medium; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Sonnet 4.6v2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning medium; tool setting standard; partial …
Task release v2026.06.24; full dataset; 500 steps; reasoning medium; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- GPT-5.5v2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning xhigh; tool setting batch tool; binary …
Task release v2026.06.24; full dataset; 150 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- GPT-5.5v2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning xhigh; tool setting batch tool; binary …
Task release v2026.06.24; full dataset; 300 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- GPT-5.5v2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary …
Task release v2026.06.24; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- GPT-5.5v2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 150 steps; reasoning xhigh; tool setting batch tool; partial…
Task release v2026.06.24; full dataset; 150 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- GPT-5.5v2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning xhigh; tool setting batch tool; partial…
Task release v2026.06.24; full dataset; 300 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- GPT-5.5v2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial…
Task release v2026.06.24; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Qwen3.7 Plusv2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning thinking; tool setting standard; binary…
Task release v2026.06.24; full dataset; 300 steps; reasoning thinking; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Qwen3.7 Plusv2026.06.24
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning thinking; tool setting standard; binary…
Task release v2026.06.24; full dataset; 500 steps; reasoning thinking; tool setting standard; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Qwen3.7 Plusv2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 300 steps; reasoning thinking; tool setting standard; partia…
Task release v2026.06.24; full dataset; 300 steps; reasoning thinking; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Qwen3.7 Plusv2026.06.24
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; reasoning thinking; tool setting standard; partia…
Task release v2026.06.24; full dataset; 500 steps; reasoning thinking; tool setting standard; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning high; tool setting batch tool; binary completi…
Task release v2.1; full dataset; 500 steps; reasoning high; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning low; tool setting batch tool; binary completio…
Task release v2.1; full dataset; 500 steps; reasoning low; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; binary completio…
Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning medium; tool setting batch tool; binary comple…
Task release v2.1; full dataset; 500 steps; reasoning medium; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary complet…
Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning high; tool setting batch tool; binary compl…
Task release v2.1; offline dataset; 500 steps; reasoning high; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning low; tool setting batch tool; binary comple…
Task release v2.1; offline dataset; 500 steps; reasoning low; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning max; tool setting batch tool; binary comple…
Task release v2.1; offline dataset; 500 steps; reasoning max; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning medium; tool setting batch tool; binary com…
Task release v2.1; offline dataset; 500 steps; reasoning medium; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary comp…
Task release v2.1; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning high; tool setting batch tool; partial credit;…
Task release v2.1; full dataset; 500 steps; reasoning high; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning low; tool setting batch tool; partial credit; …
Task release v2.1; full dataset; 500 steps; reasoning low; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; partial credit; …
Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning medium; tool setting batch tool; partial credi…
Task release v2.1; full dataset; 500 steps; reasoning medium; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit…
Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning high; tool setting batch tool; partial cred…
Task release v2.1; offline dataset; 500 steps; reasoning high; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning low; tool setting batch tool; partial credi…
Task release v2.1; offline dataset; 500 steps; reasoning low; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning max; tool setting batch tool; partial credi…
Task release v2.1; offline dataset; 500 steps; reasoning max; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning medium; tool setting batch tool; partial cr…
Task release v2.1; offline dataset; 500 steps; reasoning medium; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Claude Opus 5v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial cre…
Task release v2.1; offline dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; offline subset task count not supplied
- Muse Spark 1.2v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary complet…
Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; comparison held: MuseSpark 1.2 trajectory filename v2.1-muse_spark_1.2-1015-merged-105tasks.zip suggests a task count different from the advertised 108, actual scored denominator unverified; official flag not supplied
- Muse Spark 1.2v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit…
Task release v2.1; full dataset; 500 steps; reasoning xhigh; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; comparison held: MuseSpark 1.2 trajectory filename v2.1-muse_spark_1.2-1015-merged-105tasks.zip suggests a task count different from the advertised 108, actual scored denominator unverified; official flag not supplied
- Muse Spark 1.3v2.1
OSWorld binary completion · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; binary completio…
Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; binary completion; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; comparison held: MuseSpark 1.3 trajectory filename v2.1-muse_spark_1.3-1015-merged-106tasks.zip suggests a task count different from the advertised 108, actual scored denominator unverified; official flag not supplied
- Muse Spark 1.3v2.1
OSWorld partial credit · OSWorld (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; partial credit; …
Task release v2.1; full dataset; 500 steps; reasoning max; tool setting batch tool; partial credit; end-to-end publisher protocol; provider-specific effort labels, not equal compute; exact execution backend, runtime revision, run count, decoding configuration and result date not supplied; publisher-reported evaluation, operator not separately attested; effective task denominator unverified; comparison held: MuseSpark 1.3 trajectory filename v2.1-muse_spark_1.3-1015-merged-106tasks.zip suggests a task count different from the advertised 108, actual scored denominator unverified; official flag not supplied
- Claude Opus 5.52.1
OSWorld 2.1 · Anthropic (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Vendor-reported in Anthropic's Claude Opus 5.5 launch table; adaptive thinking at max effort unless …
Vendor-reported in Anthropic's Claude Opus 5.5 launch table; adaptive thinking at max effort unless noted; evaluated with production safeguards on, and where they intervened, cybersecurity tasks fell back to Claude Opus 4.8 and biology/frontier-LLM tasks to Claude Opus 5 (Anthropic says this likely lowers scores); partial credit scoring as labelled in the table; announcement dated 2026-09-22; measurement date unreported
- Claude Opus 4.72.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting b…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting batched tool · effort max · result date not supplied; declared publisher model Claude Opus 4.7; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 4.72.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting s…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting standard · effort max · result date not supplied; declared publisher model Claude Opus 4.7; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 4.82.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting b…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting batched tool · effort max · result date not supplied; declared publisher model Claude Opus 4.8; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 4.82.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting s…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting standard · effort max · result date not supplied; declared publisher model Claude Opus 4.8; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 52.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning high · Tool setting …
Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning high · Tool setting batch tool · effort high · result date not supplied; declared publisher model Claude Opus 5; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 52.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning low · Tool setting b…
Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning low · Tool setting batch tool · effort low · result date not supplied; declared publisher model Claude Opus 5; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 52.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning max · Tool setting b…
Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning max · Tool setting batch tool · effort max · result date not supplied; declared publisher model Claude Opus 5; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 52.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning medium · Tool settin…
Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning medium · Tool setting batch tool · effort medium · result date not supplied; declared publisher model Claude Opus 5; exact execution backend, run count and decoding configuration not supplied
- Claude Opus 52.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning xhigh · Tool setting…
Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning xhigh · Tool setting batch tool · effort xhigh · result date not supplied; declared publisher model Claude Opus 5; exact execution backend, run count and decoding configuration not supplied
- Claude Sonnet 4.62.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting s…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning max · Tool setting standard · effort max · result date not supplied; declared publisher model Claude Sonnet 4.6; exact execution backend, run count and decoding configuration not supplied
- Claude Sonnet 4.62.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning medium · Tool settin…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning medium · Tool setting standard · effort medium · result date not supplied; declared publisher model Claude Sonnet 4.6; exact execution backend, run count and decoding configuration not supplied
- GPT-5.52.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning xhigh · Tool setting…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning xhigh · Tool setting batch tool · effort xhigh · result date not supplied; declared publisher model GPT-5.5; exact execution backend, run count and decoding configuration not supplied
- GPT-5.6 Sol2.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning max · Tool setting b…
Task release v2026.08.08; full dataset; 500 steps; binary completion; Reasoning max · Tool setting batch tool · effort max · result date not supplied; declared publisher model GPT-5.6 Sol; exact execution backend, run count and decoding configuration not supplied
- Kimi K2.62.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning enabled · Tool setti…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning enabled · Tool setting standard · result date not supplied; declared publisher model Kimi 2.6; exact execution backend, run count and decoding configuration not supplied
- MiniMax M32.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning enabled · Tool setti…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning enabled · Tool setting standard · result date not supplied; declared publisher model MiniMax M3; exact execution backend, run count and decoding configuration not supplied
- Qwen3.7 Plus2.0
OSWorld 2.0 · OSWorld via Epoch AI (reviewed publisher results)
Single-model result · Positions withheld
Measurement date not reported
Scope · Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning thinking · Tool sett…
Task release v2026.06.24; full dataset; 500 steps; binary completion; Reasoning thinking · Tool setting standard · result date not supplied; declared publisher model Qwen 3.7-Plus; exact execution backend, run count and decoding configuration not supplied