Artificial Analysis
2 Tools · 299 Intel
Artificial Analysis is an independent AI benchmarking organization, founded in 2023 by Micah Hill-Smith and George Cameron, that runs its own evaluations of AI models and inference providers rather than relying on self-reported figures from AI labs. Its free public leaderboards and Intelligence Index have become widely referenced industry benchmarks for comparing model quality, speed, and cost. The company is backed by investors including Nat Friedman, Daniel Gross, and Andrew Ng.
Is this you? Sign in with X to claim this profile.
Tools
Independent benchmarks comparing AI models and providers on intelligence, coding, speed, latency, and price.
Artificial Analysis's platform for creating and running custom benchmarks against AI models on your own tasks.
Intel
Artificial Analysis benchmarks Nanbeige4.3-33B, a dense Apache 2.0 open-weights reasoning model: tied with Qwen3.8 27B at 20, better terminal use and abstention, weaker knowledge, and higher token consumption.
Thinking Machines releases Inkling, a 975B (41B active) multimodal open-weights model scoring 41 on the Artificial Analysis index, leading US open models and using fewer output tokens than GLM-5.2 and Kimi K2.6.
Artificial Analysis benchmarks NVIDIA's Nemotron 3.5 Super VL and Bolt 74B-A9B: index scores, token efficiency, ~630 tokens/s output speed and NVFP4 accuracy retention against BF16.
Ant Group releases Ling-3.0-flash-Fin, an MIT-licensed 124B MoE (5.1B active) finance model with a 256K context. It scores 23 on the Intelligence Index and 24 on Finance & Accounting but struggles on agentic tasks.
Artificial Analysis launches the Endpoint Accuracy Index, comparing serverless endpoints to a self-hosted reference on tool calling, reasoning and long context for GLM-5.2, gpt-oss-120b and DeepSeek V4 Pro.
Artificial Analysis benchmarks Google's Gemini 3.8 Live speech-to-speech model. The Extended Thinking (High) variant leads the Speech to Speech Index at 82.6 and Tau Voice at 68.6%, with latency of 1.35s to first audio and lower cost than GPT-Live-1 and Grok Voice.
Kimi K3 (2.8T parameters) scores 57 on the Artificial Analysis index, tops AutomationBench-AA and uses 21% fewer output tokens than K2.6, at higher pricing. Weights are promised but not yet released.
A performance and pricing comparison of 15 API providers serving the GLM-5.2 (max) model, measuring output speed, time-to-first-token latency, and blended token pricing. Databricks leads on both speed (394.6 t/s) and latency (6.08s), while DeepInfra and CoreWeave offer the cheapest rates at $0.49 per million tokens, with up to a 757% speed gap and 5.2x price gap across providers.
Artificial Analysis reviews the three leading open-weights models, all trillion-plus-parameter MoEs with permissive licenses, quantifying gaps to GPT-5.5 and Gemini on hard evals and their dominance of the price frontier.
Artificial Analysis relaunches its Coding Agent Index with Cognition's FrontierCode v1.1 and stricter repo Q&A grading. Claude Code with Opus 5.5 leads at 61.7, scores drop about 7 points versus v1.5, and a cost frontier shows per-task pricing.
Artificial Analysis reports that MiMo-V2.6-Pro, Claude Opus 5.5, GPT-6 Luna, and GPT-6 Sol added eleven points to its Intelligence Index cost-per-task Pareto frontier, with Opus 5.5 the top scorer at 58 for $5.98 per task.
Artificial Analysis releases Capability Indices v1.1, adding agentic tool use from AutomationBench-AA, AA-Briefcase and GDP.pdf and retuning domain weights for finance, legal, ops, healthcare, engineering and economics.
Artificial Analysis benchmarks Mistral Large 4 Preview: 38 on the Intelligence Index, 50 on the Cyber Index with 82% on CyberGym-E2E, and a cost per task above 4x comparable open weights models. It has a 512k context window, and weights are due at the end of October.
Artificial Analysis finds MiniMax M2.7 scores 50, up 8 points in a month, with big agentic and hallucination gains and index cost of $176 versus $547 for GLM-5.
On the Artificial Analysis Coding Agent Index, Meta's Muse Spark 1.3 (xhigh) scores 64 in Muse Code at $1.72 per task, the cheapest agent above 60. The partner-only max variant scores 68, level with Claude Opus 5 (xhigh), using about a third fewer tokens per task.
Artificial Analysis ranks Kimi K3.1 Pro first among openly available models on its Cyber Index at 64%, with a record 95% on CyberGym-E2E-AA. It costs $2.81 per task and meets no refusals, while trusted-access GPT-6 Sol scores 69%.
Artificial Analysis tests all four GPT-5 reasoning efforts: High scores 68 on its index, Minimal is near GPT-4.1, token usage varies 23x, and GPT-5 tops long-context reasoning on AA-LCR.
Artificial Analysis benchmarks Meta's Muse Wonder (52) and Wonder Full (56) against Muse Spark 1.3 (48). Gains are largest in agentic work, API pricing is $4/$20, and cost per task reaches $4.52.
Artificial Analysis benchmarks Meta's Muse Glimmer, a 30B dense open-weights model under Apache 2.0. It runs on a single GPU with hybrid attention, does well on tool use, and lags on agentic knowledge work and hallucination. The score reported on the thread differs from the model page.
Artificial Analysis tracks the intelligence-versus-cost frontier for frontier models. Its Index v5 puts 60% of the weight on private benchmarks, and its Coding Agent Index compares agents on cost per task, with top models differing by more than 13x in cost.
Artificial Analysis finds GPT-5.5 tops its index by 3 points, doubles per-token pricing yet nets about 20% higher index cost, and offers an effort ladder, with high accuracy but a high hallucination rate.
Artificial Analysis benchmarks Grok Voice 3.0 as the top Speech to Speech model at 86.9. It leads on task success and audio reasoning, trails GPT-Live-1 on Tau Voice and Gemini Live on preference, and costs $4.80 per input hour.
Artificial Analysis benchmarked Gemini 3.7 Flash: 56 on its Intelligence Index (+4 over 3.6 Flash), about 340 tok/s, $0.75/$3.75 per 1M tokens under year-end discounted pricing, $0.40 per task, and top scores on AutomationBench-AA and AA-AnalystAgent.
Artificial Analysis benchmarks GLM-5.2: 51 on its Intelligence Index, 11 points over GLM-5.1 at the same 744B size, with gains on scientific reasoning and TerminalBench, and pricing unchanged from the prior version.