Model results

BrokenArXiv

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.

Official source

← All evaluation groups

BrokenArXiv August 2026

Source: MathArena (reviewed publisher results)
Version: 2026-08-native-harness
Measurement date: Not reported
Metric: % · Higher is better
Evaluation scope: Mean rubric points normalized to percent; 56 questions; 0–3 grading, full credit requires identifying the statement as false; not binary accuracy; August 2026 dataset; harness=Antigravity CLI; configuration=Gemini 3.8 Flash; Python and SageMath tools; no internet; budget/time-limit instructions varied across models, see methodology https://matharena.ai/arxiv_august/; 95% CI 6.74%; cost per problem $0.60; result date not supplied; captured 2026-09-17; publisher flags model release after competition release

Fewer than two comparable models. No ranking is assigned. Missing metadata is shown as unreported; matching imported metadata is not independent verification of the protocol.

Gemini 3.8 Flash
google/gemini-3.8-flash-20260902
27.38%
View modelEvidence independent