AI MODEL INDEX
Latest AI model performance and cost
Compare an independent intelligence index with cost per benchmark task, including token usage.
Snapshot:
Latest model intelligence index
Representative reasoning setting per model · higher is better
Selected model detailsClaude Opus 5.5Anthropic · max · default fallbackModel source
- Index
- 57.6
- Per task
- $5.98
- Output speed
- 95.9 t/s
- Context
- 1M
Intelligence vs cost per task
Models closer to the upper-left deliver more intelligence at lower task cost
Selected model detailsClaude Opus 5.5Anthropic · max · default fallbackModel source
- Index
- 57.6
- Per task
- $5.98
- Output speed
- 95.9 t/s
- Context
- 1M
| Model | Setting | Index | Cost/task | Output speed | Context |
|---|---|---|---|---|---|
| Claude Opus 5.5Anthropic | max · default fallback | 57.6 | $5.98 | 95.9 t/s | 1M |
| Claude Sonnet 5.5Anthropic | max · default fallback | 56 | $5.46 | 133.8 t/s | 1M |
| Claude Fable 5.1Anthropic | max · default fallback | 53.4 | $7.63 | 68.5 t/s | 1M |
| GPT-6 AstraOpenAI | max | 52.7 | $3.26 | 56.3 t/s | 1M |
| Gemini 4 ArgonGoogle | high | 52.6 | $1.99 | 81.5 t/s | 1M |
| Muse WonderMetaOfficial availability: Model API and Muse Code are currently unavailable in South Korea. This is not Wonder Full. | max | 52.4 | $4.52 | 145.1 t/s | 1M |
| GPT-6.1 SolOpenAI | max | 51.8 | $0.72 | 54.3 t/s | 1M |
| Grok 4.7SpaceXAI | xhigh | 46.4 | $3.74 | 88.9 t/s | 500K |
| MiMo-V2.6-ProXiaomi | provider default · no explicit effort field | 46.3 | $0.13 | 52.6 t/s | 1M |
| DeepSeek V4.1 Flash 1017DeepSeek | max | 45.1 | $0.26 | 203.7 t/s | 1M |
| GLM-5.3Z AI | max | 44.8 | $2.01 | 76.7 t/s | 1M |
| Kimi K3Kimi | max | 43.6 | $2.00 | 36.9 t/s | 1M |
| Claude Haiku 5.5Anthropic | max | 43.4 | $0.21 | 238.3 t/s | 1M |
| Gemini 3.8 FlashGoogle | high | 40.9 | $1.24 | 141.5 t/s | 1M |
| GPT-6 LunaOpenAI | max | 38.1 | $0.07 | 135.2 t/s | 1M |
Scope, reasoning settings and cost methodology
This is a dated selection of models and representative configurations, not an exhaustive market ranking. Artificial Analysis v4.3.2 is primarily an English/text benchmark, not an overall score for Korean, image or voice quality.
max, high and xhigh identify model-specific evaluated configurations, not equal reasoning budgets. USD/task is a weighted benchmark cost reflecting cache and reasoning-token usage, not an API list price per million tokens. Do not interpret differences from the earlier v4.3 snapshot as performance gains under identical conditions.