Public benchmarks
This page compares only the Artificial Analysis Intelligence Index, preserving the leaderboard’s original scores and reasoning settings. It is a third-party index, not a percentage or an Anran score. Data checked on 2026-10-21.
The table uses one source: Artificial Analysis Intelligence Index. Scores preserve the published model version and reasoning setting; they are index values, not percentages or house ratings.
| Provider / model | Artificial Analysis Intelligence Index | Reasoning setting | Checked | Official sources |
|---|---|---|---|---|
| OpenAI · GPT-6 Astra | 53 | max | 2026-10-21 | Artificial Analysis leaderboard |
| OpenAI · GPT-6.1 Sol | 52 | max | 2026-10-21 | Artificial Analysis leaderboard |
| Anthropic · Claude Opus 5.5 | 58 | max with fallback | 2026-10-21 | Artificial Analysis leaderboard |
| Anthropic · Claude Fable 5.1 | 53 | max with fallback | 2026-10-21 | Artificial Analysis leaderboard |
| Anthropic · Claude Sonnet 5.5 | 56 | max with fallback | 2026-10-21 | Artificial Analysis leaderboard |
| Anthropic · Claude Haiku 5.5 | 43 | max | 2026-10-21 | Artificial Analysis leaderboard |
| Google · Gemini 4 Argon | 53 | high | 2026-10-21 | Artificial Analysis leaderboard |
| DeepSeek · DeepSeek-V4.1 Flash | 45 | max | 2026-10-21 | Artificial Analysis leaderboard |
| Alibaba Cloud Qwen · Qwen3.8 Max | 45 | Not specified by leaderboard | 2026-10-21 | Artificial Analysis leaderboard |
| xAI · Grok 4.7 | 46 | xhigh | 2026-10-21 | Artificial Analysis leaderboard |
How to read the scores
Section titled “How to read the scores”- Artificial Analysis uses the Intelligence Index to compare model capability; the same model can receive different results at different reasoning-effort settings.
- This guide preserves the model versions and modes shown on the leaderboard, such as
max,high,xhigh, andmax with fallback. The Qwen3.8 Max row does not specify a reasoning setting, so it is labeled “Not specified by leaderboard.” - The benchmark method covers quality, performance, and price to estimate end-to-end user experience. Its methodology also explains token units, task costs, and related measures.
- The leaderboard changes. Scores from different dates or model settings should not be treated as a fixed ranking; check the original leaderboard for the latest entries.
Sources: Artificial Analysis model leaderboard · Benchmark methodology