Benchmarks
What each test measures, and what it cannot tell you.
Three numbers per model
118 models, sorted by intelligence. The count beside each score is how many of the nine indices that model carries, so a model measured on four is marked as such.
| Model | Intelligence v5, 9 Nov 2026 | Speed tok/s, 9 Nov 2026 | Cost $/1M in / out, 9 Nov 2026 |
|---|---|---|---|
| claude-fable-5-2 | 64.6 · 7/9 | 95 tok/s | $8.00 / $40.00 |
| claude-fable-5-2-xhigh | 62.2 · 6/9 | 83 tok/s | $8.00 / $40.00 |
| claude-fable-5-2-high | 61.8 · 6/9 | 76 tok/s | $8.00 / $40.00 |
| gemini-4-carbon | 60.0 · 6/9 | 100 tok/s | $4.00 / $20.00 |
| claude-opus-5-5 | 59.1 · 7/9 | 94 tok/s | $4.00 / $20.00 |
| gemini-4-carbon-medium | 58.3 · 6/9 | 96 tok/s | $4.00 / $20.00 |
| claude-opus-5-5-xhigh | 57.4 · 7/9 | 81 tok/s | $4.00 / $20.00 |
| gpt-6-astra | 56.9 · 7/9 | 49 tok/s | $10.00 / $50.00 |
| claude-fable-5-2-medium | 56.4 · 6/9 | 75 tok/s | $8.00 / $40.00 |
| gpt-6-astra-xhigh | 55.3 · 7/9 | 47 tok/s | $10.00 / $50.00 |
| gpt-6-astra-high | 54.8 · 7/9 | 45 tok/s | $10.00 / $50.00 |
| gpt-6-1-sol | 54.6 · 7/9 | 69 tok/s | $2.00 / $10.00 |
| claude-sonnet-5-5 | 54.5 · 7/9 | 128 tok/s | $2.00 / $10.00 |
| gemini-4-carbon-low | 54.2 · 6/9 | 89 tok/s | $4.00 / $20.00 |
| claude-fable-5-2-low | 53.9 · 6/9 | 75 tok/s | $8.00 / $40.00 |
| gpt-6-1-sol-xhigh | 52.5 · 7/9 | 69 tok/s | $2.00 / $10.00 |
| claude-opus-5-5-high | 51.8 · 7/9 | 78 tok/s | $4.00 / $20.00 |
| gpt-6-astra-medium | 51.6 · 7/9 | 42 tok/s | $10.00 / $50.00 |
| gpt-6-1-terra | 51.3 · 7/9 | 103 tok/s | $1.00 / $5.00 |
| gemini-4-argon | 50.4 · 7/9 | 89 tok/s | $4.00 / $20.00 |
| gpt-6-1-sol-high | 50.3 · 7/9 | 61 tok/s | $2.00 / $10.00 |
| muse-wonder-full | 50.1 · 7/9 | — | — / — |
| claude-fable-5-1-xhigh | 50.0 · 7/9 | 59 tok/s | $10.00 / $50.00 |
| claude-sonnet-5-5-xhigh | 49.9 · 7/9 | 103 tok/s | $2.00 / $10.00 |
| claude-fable-5-1 | 49.2 · 7/9 | 68 tok/s | $10.00 / $50.00 |
| claude-opus-5-5-medium | 48.2 · 7/9 | 77 tok/s | $4.00 / $20.00 |
| claude-fable-5-1-high | 47.2 · 7/9 | 57 tok/s | $10.00 / $50.00 |
| kimi-k3-1-pro | 46.5 · 7/9 | 29 tok/s | $2.00 / $10.00 |
| gpt-6-1-sol-medium | 46.3 · 7/9 | 55 tok/s | $2.00 / $10.00 |
| gpt-6-1-terra-xhigh | 45.3 · 6/9 | 91 tok/s | $1.00 / $5.00 |
| claude-opus-5 | 44.2 · 7/9 | 52 tok/s | $5.00 / $25.00 |
| gpt-6-sol | 44.2 · 7/9 | 154 tok/s | $2.00 / $10.00 |
| muse-wonder | 43.6 · 7/9 | 137 tok/s | $4.00 / $20.00 |
| gpt-6-astra-low | 43.3 · 7/9 | 42 tok/s | $10.00 / $50.00 |
| claude-fable-5-1-medium | 43.2 · 7/9 | 56 tok/s | $10.00 / $50.00 |
| gpt-6-1-terra-high | 42.4 · 6/9 | 89 tok/s | $1.00 / $5.00 |
| claude-sonnet-5-5-high | 41.5 · 7/9 | 105 tok/s | $2.00 / $10.00 |
| muse-wonder-high | 40.5 · 6/9 | 151 tok/s | $4.00 / $20.00 |
| claude-fable-5-1-low | 38.7 · 7/9 | 55 tok/s | $10.00 / $50.00 |
| gpt-6-1-terra-medium | 38.2 · 6/9 | 88 tok/s | $1.00 / $5.00 |
| muse-spark-1-3 | 36.6 · 7/9 | 173 tok/s | $1.25 / $4.25 |
| claude-haiku-5-5 | 36.5 · 7/9 | 232 tok/s | $0.10 / $0.50 |
| muse-ember | 36.0 · 7/9 | 542 tok/s | $0.10 / $0.50 |
| grok-4-7 | 35.5 · 7/9 | 69 tok/s | $2.00 / $6.00 |
| grok-4-7-high | 35.2 · 7/9 | 74 tok/s | $2.00 / $6.00 |
| claude-opus-5-5-low | 34.8 · 7/9 | 75 tok/s | $4.00 / $20.00 |
| gpt-6-1-sol-low | 34.8 · 7/9 | 56 tok/s | $2.00 / $10.00 |
| claude-haiku-5-5-xhigh | 34.6 · 7/9 | 186 tok/s | $0.10 / $0.50 |
| tencent-hy4 | 34.6 · 6/9 | — | $0.83 / $2.50 |
| muse-ember-xhigh | 34.1 · 6/9 | 512 tok/s | $0.10 / $0.50 |
| qwen3-8-max | 33.7 · 7/9 | 40 tok/s | $2.00 / $6.00 |
| deepseek-v4-1-flash-1017 | 32.6 · 7/9 | 144 tok/s | $0.30 / $1.20 |
| gpt-5-6-terra | 32.3 · 7/9 | 130 tok/s | $2.00 / $12.00 |
| glm-5-3 | 31.7 · 7/9 | 85 tok/s | $1.40 / $4.40 |
| grok-4-7-low | 31.1 · 7/9 | 56 tok/s | $2.00 / $6.00 |
| grok-4-6-xhigh | 31.0 · 7/9 | 73 tok/s | $2.00 / $6.00 |
| claude-sonnet-5-5-medium | 30.8 · 7/9 | 101 tok/s | $2.00 / $10.00 |
| gemini-3-8-flash | 30.8 · 7/9 | 126 tok/s | $0.75 / $3.75 |
| inkling-2-small | 30.6 · 7/9 | 111 tok/s | $0.30 / $1.00 |
| gpt-6-1-terra-non-reasoning | 30.5 · 6/9 | 87 tok/s | $1.00 / $5.00 |
| gemini-3-8-flash-medium | 30.1 · 7/9 | 123 tok/s | $0.75 / $3.75 |
| claude-haiku-5-5-high | 29.9 · 7/9 | 180 tok/s | $0.10 / $0.50 |
| mimo-v2-6-pro | 29.9 · 7/9 | 58 tok/s | $0.43 / $0.87 |
| ling-3-1-flash-1019 | 29.6 · 7/9 | — | $0.30 / $0.90 |
| gpt-6-1-terra-low | 29.4 · 6/9 | 87 tok/s | $1.00 / $5.00 |
| minimax-m3-1-flash | 29.1 · 7/9 | 166 tok/s | $0.15 / $0.60 |
| deepseek-v4-1-flash-1017-high | 28.6 · 6/9 | 154 tok/s | $0.30 / $1.20 |
| step-5 | 28.5 · 7/9 | 165 tok/s | $1.00 / $2.70 |
| gpt-6-luna | 27.9 · 7/9 | 125 tok/s | $0.10 / $0.50 |
| glm-5-3-flash | 27.8 · 7/9 | 57 tok/s | $0.15 / $0.50 |
| mimo-v2-6-flash | 26.7 · 7/9 | 41 tok/s | $0.14 / $0.28 |
| mistral-large-4 | 26.1 · 7/9 | 133 tok/s | $0.70 / $2.10 |
| claude-sonnet-5-5-low | 25.6 · 7/9 | 99 tok/s | $2.00 / $10.00 |
| mistral-large-4-preview | 25.3 · 7/9 | 138 tok/s | $0.70 / $2.10 |
| muse-spark-open | 24.9 · 7/9 | 233 tok/s | $1.25 / $4.25 |
| claude-haiku-5-5-medium | 24.6 · 7/9 | 173 tok/s | $0.10 / $0.50 |
| agnes-2-5-pro-theta | 24.0 · 6/9 | 251 tok/s | $0.10 / $0.30 |
| gpt-5-6-luna | 23.8 · 7/9 | 127 tok/s | $0.20 / $1.20 |
| gpt-6-luna-xhigh | 23.5 · 7/9 | 120 tok/s | $0.10 / $0.50 |
| deepseek-v4-pro | 23.1 · 6/9 | 82 tok/s | $1.32 / $3.96 |
| glm-5-3-low | 23.1 · 7/9 | 78 tok/s | $1.40 / $4.40 |
| mistral-large-4-medium | 23.1 · 6/9 | 136 tok/s | $0.70 / $2.10 |
| gemini-3-8-flash-low | 22.5 · 7/9 | 120 tok/s | $0.75 / $3.75 |
| gpt-6-luna-high | 21.4 · 7/9 | 118 tok/s | $0.10 / $0.50 |
| nanbeige4-3-33b | 20.4 · 6/9 | — | — / — |
| qwen3-8-27b | 20.1 · 7/9 | 43 tok/s | $0.50 / $3.00 |
| minimax-m3 | 18.9 · 7/9 | 137 tok/s | $0.30 / $1.20 |
| claude-haiku-5-5-low | 18.6 · 7/9 | 164 tok/s | $0.10 / $0.50 |
| deepseek-v4-1-flash-1017-non-reasoning | 18.5 · 6/9 | 146 tok/s | $0.30 / $1.20 |
| gpt-6-luna-medium | 18.5 · 7/9 | 114 tok/s | $0.10 / $0.50 |
| k2-horizon-375b-a23b | 17.8 · 7/9 | 916 tok/s | $0 / $0 |
| reflection-beam | 17.7 · 7/9 | 389 tok/s | $0.40 / $1.20 |
| qwen3-8-27b-medium | 17.5 · 7/9 | 46 tok/s | $0.50 / $3.00 |
| kimi-k3-low | 17.4 · 7/9 | 47 tok/s | $3.00 / $15.00 |
| qwen3-8-27b-low | 15.7 · 7/9 | 46 tok/s | $0.50 / $3.00 |
| nemotron-3-5-super-vl | 13.6 · 7/9 | 296 tok/s | $0.17 / $0.50 |
| inkling-small | 12.9 · 7/9 | 96 tok/s | $0.30 / $1.20 |
| deepseek-v4-pro-non-reasoning | 12.6 · 6/9 | 75 tok/s | $1.32 / $3.96 |
| inkling | 12.5 · 7/9 | 193 tok/s | $1.00 / $4.05 |
| nvidia-nemotron-3-ultra-550b-a55b | 11.8 · 7/9 | 250 tok/s | $0.60 / $2.70 |
| gemini-3-5-flash-lite | 11.7 · 7/9 | 400 tok/s | $0.30 / $2.50 |
| qwen3-8-27b-non-reasoning | 10.0 · 7/9 | 48 tok/s | $0.50 / $3.00 |
| gpt-6-luna-low | 9.7 · 7/9 | 118 tok/s | $0.10 / $0.50 |
| gpt-6-luna-non-reasoning | 9.4 · 7/9 | 113 tok/s | $0.10 / $0.50 |
| k-exaone-2-0-0803 | 7.6 · 6/9 | — | — / — |
| claude-4-5-haiku-reasoning | 7.4 · 7/9 | 88 tok/s | $1.00 / $5.00 |
| muse-glimmer | 6.3 · 6/9 | 107 tok/s | $0.30 / $1.20 |
| command-a-plus | 6.2 · 7/9 | 237 tok/s | $0 / $0 |
| gemini-3-1-flash-lite-preview | 6.2 · 7/9 | 353 tok/s | $0.25 / $1.50 |
| nemotron-3-5-lightning | 5.5 · 7/9 | 274 tok/s | $0.06 / $0.20 |
| granite-4-2-8b | 5.1 · 6/9 | 140 tok/s | $0.06 / $0.25 |
| granite-4-2-3b | 4.7 · 6/9 | — | — / — |
| mistral-medium-3-5 | 4.6 · 7/9 | 166 tok/s | $1.50 / $7.50 |
| gemma-4-31b | 4.4 · 6/9 | 35 tok/s | $0 / $0 |
| mistral-small-4 | 4.3 · 7/9 | 171 tok/s | $0.15 / $0.60 |
| gpt-oss-120b | 3.8 · 7/9 | 210 tok/s | $0.15 / $0.60 |
| north-mini-code | 3.4 · 7/9 | 176 tok/s | $0 / $0 |
| nvidia-nemotron-3-nano-4b | 1.8 · 6/9 | — | — / — |
Intelligence is Artificial Analysis's own index: one number per model, already a composite of its nine domain indices. It is not re-averaged here, because a composite of a composite is a number nobody can check.
Index version 5, read 9 November 2026. Data: Artificial Analysis (artificialanalysis.ai)
Every reading we have recorded, one chart per index.
What each test measures
-
Why benchmark scores do not predict your workload
The top three models differ by two points and the error bars are wider than that. Saturation, contamination and the harness explain the rest.
-
SWE-bench, LiveCodeBench and what each one measures
SWE-bench asks whether a patch makes real tests pass. LiveCodeBench asks whether a model can solve a problem published after it was trained. Neither measures your job.
-
How to run your own eval in an afternoon
Fifty real inputs, one check that runs, and a scoreboard you keep. That is the whole method, and it beats every public leaderboard for your decision.