Frontier models, benchmarked continuously.
Quality, cost and speed for the leading AI models — refreshed on a live cycle · last updated 2026-09-19.
Intelligence, coding & agentic indexes
Composite indexes by Artificial Analysis · price per 1M tokens (input / output)
| # | Model | Intelligence | Coding | Agentic | $ / 1M in | $ / 1M out |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) | 53.4 | 81.6 | 58 | $5.00 | $25.00 |
| 2 | Qwen3.8 Max | 53.4 | 68.9 | 49.9 | — | — |
| 3 | GPT-6 Astra (max) | 52.8 | 76.9 | 51.5 | $11.00 | $55.00 |
| 4 | Claude Opus 5 (Adaptive Reasoning, Max Effort) | 50.7 | 78 | 56.2 | $5.50 | $27.50 |
| 5 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | 49.7 | 76.5 | 51 | $10.00 | $50.00 |
| 6 | GPT-5.6 Sol (max) | 47.1 | 77.4 | 50.5 | $4.00 | $20.00 |
| 7 | Qwen3.8 Max (0902) | 45.4 | 76.2 | 56.1 | $2.00 | $6.00 |
| 8 | GLM-5.3 (max) | 44.9 | 74.8 | 53.4 | $0.890 | $2.80 |
| 9 | Grok 4.6 (high) | 44.4 | 76.8 | 53.4 | $4.00 | $12.00 |
| 10 | Kimi K3 (max) | 43.8 | 76.2 | 50.6 | $2.10 | $10.95 |
| 11 | GPT-5.6 Terra (max) | 42.3 | 76.7 | 43.7 | $4.00 | $24.00 |
| 12 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | 42 | 74.3 | 42.6 | $10.00 | $50.00 |
| 13 | GLM-5.3-Flash | 41.9 | 71.5 | 51.2 | $0.150 | $0.500 |
| 14 | Gemini 3.8 Flash (high) | 41.2 | 76.3 | 41.1 | $1.35 | $6.75 |
| 15 | Qwen3.8 2.4T A95B | 40 | 71.9 | 50.4 | $2.00 | $6.00 |
| 16 | Muse Spark 1.2 (xhigh) | 39.8 | 72.2 | 44 | $1.25 | $4.25 |
| 17 | DeepSeek V4.1 Flash (Reasoning, Max Effort) | 39.5 | — | — | $0.200 | $0.600 |
| 18 | Gemini 3.7 Flash (high) | 39.4 | 76.1 | 36.4 | $1.35 | $6.75 |
| 19 | Grok 4.5 (high) | 39.1 | 72.4 | 42.1 | $4.00 | $12.00 |
| 20 | GPT-5.5 (xhigh) | 38.6 | 74.9 | 37.3 | $2.50 | $15.00 |
| 21 | Claude Sonnet 5 (Adaptive Reasoning, Max Effort) | 38.4 | 71.5 | 44.3 | $2.20 | $11.00 |
| 22 | GPT-5.6 Luna (max) | 37.5 | 71.4 | 42.7 | $0.400 | $2.40 |
| 23 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) | 36.3 | 68.8 | 42.3 | $1.32 | $3.96 |
| 24 | DeepSeek V4 Flash 0731 (Reasoning, Max Effort) | 34.5 | 69.1 | 41.7 | $0 | $0 |
| 25 | Muse Spark 1.1 (xhigh) | 34.3 | 71.3 | 27.5 | $1.25 | $4.25 |
| 26 | Gemini 3.6 Flash (high) | 34.3 | 69.2 | 30.2 | $1.35 | $6.75 |
| 27 | GLM-5.2 (max) | 34 | 68.8 | 39.4 | $2.25 | $8.00 |
| 28 | Qwen3.8 27B (xhigh) | 33.9 | 68.1 | 46.5 | $0 | $0 |
| 29 | Gemini 3.5 Flash (high) | 33 | 70.1 | 27.3 | $2.70 | $16.20 |
| 30 | DeepSeek V4 Pro (Reasoning, Max Effort) | 30.9 | 59.4 | 27.7 | $1.15 | $2.55 |
Reasoning, agents & search — re-run on live models
GPQA Diamond
Graduate-level science reasoning
| # | Model | Score | $ / task |
|---|---|---|---|
| 1 | Fugu Ultra | 94.6% | $1.15 |
| 2 | Gemini 3.1 Pro Preview | 94.4% | $0.202 |
| 3 | GPT-6 Astra | 94.4% | $0.099 |
| 4 | Gemini 3.7 Flash | 94.3% | $0.031 |
| 5 | GPT-5.5 | 93.8% | $0.308 |
| 6 | GPT-5.6 Sol Pro | 93.8% | $0.445 |
| 7 | Fugu Ultra V2 | 93.3% | $0.503 |
| 8 | Grok 4.6 | 93.3% | $0.122 |
| 9 | Gemini 3.6 Flash | 92.8% | $0.067 |
| 10 | Gemini 3.5 Flash | 92.8% | $0.142 |
| 11 | GPT-5.6 Sol | 91.9% | $0.071 |
| 12 | Kimi K3 | 91.9% | $0.139 |
| 13 | Union Alpha | 90.9% | — |
| 14 | Auto Router | 90.7% | $0.0079 |
| 15 | MiniMax M3 | 90.6% | $0.030 |
No single model wins every benchmark.
The leader changes by task: one model tops reasoning, another tops coding, a third wins agentic search. That is exactly why MultipleChat exists — put the leaders side by side, let them draft, challenge and verify each other, and keep the best answer.
Quality, cost and speed together
A benchmark score without a price tag is marketing. Every leaderboard here pairs quality with cost per task or per million tokens.
Live & open
Refreshed continuously and downloadable as JSON. Our own quarterly comparison study lives at the MultipleChat AI Benchmark.
Attribution
Source: OpenRouter (openrouter.ai/rankings) incl. Artificial Analysis indexes, as of 2026-09-19T14:13:10Z. Licensed under CC BY 4.0.