We tested 136 models across 44 benchmarks on success rates, latency and cost per task.
Each point is a model. The dashed line joins the models that no other model beats on both cost and score.
# | Model | License | AIM score | Input $/1M | Output $/1M | Cost per task | TTFT | Context | |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Gemini 4 CarbonGoogle | Proprietary | 85 | $1.00 | $5.00 | $0.04 | 5.75 s | 1M | |
| 2 | Claude Opus 5.5Anthropic | Proprietary | 78 | $4.00 | $20.00 | $2.03 | 3.06 s | 1M | |
| 3 | Claude Fable 5.2Anthropic | Proprietary | 77 | $8.00 | $40.00 | $0.15 | 5.07 s | 1M | |
| 4 | Gemini 4 ArgonGoogle | Proprietary | 75 | $2.00 | $10.00 | $1.23 | 6.84 s | 1M | |
| 5 | GPT-6 AstraOpenAI | Proprietary | 74 | $10.00 | $50.00 | $1.42 | 6.68 s | 1.1M | |
| 6 | GPT-6.1 SolOpenAI | Proprietary | 74 | $2.00 | $10.00 | $0.27 | 4.07 s | 1.1M | |
| 7 | Claude Sonnet 5.5Anthropic | Proprietary | 74 | $2.00 | $10.00 | $2.13 | 1.31 s | 1M | |
| 8 | Claude Fable 5.1Anthropic | Proprietary | 74 | $10.00 | $50.00 | $3.76 | 3.11 s | 1M | |
| 9 | GPT-6.1 TerraOpenAI | Proprietary | 71 | $1.00 | $5.00 | $0.23 | 1.77 s | 1.1M | |
| 10 | GPT-6 SolOpenAI | Proprietary | 70 | $2.00 | $10.00 | $0.56 | 2.45 s | 1.1M |
AIMultiple benchmarks run in-house, alongside public results reported by providers or third-party leaderboards. Best result per column is highlighted.
Model | AIM score | AIM-Agentic-RAG | AIM-FinanceReasoning | AIM-Text-to-SQL | AIM-AI-Bias | Terminal-Bench 4.0 | CritPt | MMMU-Pro | SciCode | FrontierMath |
|---|---|---|---|---|---|---|---|---|---|---|
| Gemini 4 Carbon | 85 | 92 | 90 | 78 | - | 65 | - | 89 | - | 95 |
| Claude Opus 5.5 | 78 | 90 | 92 | 88 | 94 | 60 | 32 | 88 | 67 | 95 |
| Claude Fable 5.2 | 77 | 97 | 92 | 86 | - | 62 | 36 | - | 68 | 100 |
| Gemini 4 Argon | 75 | 94 | 91 | 79 | 98 | 57 | 27 | 88 | 62 | 78 |
| GPT-6 Astra | 74 | 84 | 87 | 66 | 98 | 60 | 32 | 87 | 56 | 98 |
| GPT-6.1 Sol | 74 | 81 | 90 | 65 | 98 | 56 | 32 | 86 | 56 | 100 |
| Claude Sonnet 5.5 | 74 | 85 | 89 | 74 | 93 | 64 | 31 | 86 | 61 | 80 |
| Claude Fable 5.1 | 74 | 84 | 90 | 67 | 98 | 55 | 31 | 86 | 63 | 88 |
| GPT-6.1 Terra | 71 | 84 | 87 | 57 | 95 | 56 | 28 | 85 | 55 | 93 |
| GPT-6 Sol | 70 | 82 | 89 | 54 | 96 | 44 | 31 | 83 | 58 | 90 |
Swipe sideways to see all benchmarks
Models are ranked by the AIM LLM Index, a weighted average of AIMultiple's own benchmarks and public benchmarks on coding, reasoning, finance and agentic tasks. Prices are the lowest list API price across providers. Cost per task comes from each benchmark's token counts or published costs; the default view averages it over the benchmarks a model has a cost on.
Read the full methodologyHand-graded tasks on coding, reasoning, finance and agentic workflows, re-run as new models are released.
Explore AIM LLM Index