Language models ranked by intelligence and task cost
Read the Intelligence Index next to evaluation cost, then choose a reasoning setting that fits the US workflow you actually run.

👁 112 reads
On a timed help desk, the wait before the first word matters. For a night job comparing long policies, I care more about the number of correct comparisons I can buy. The Artificial Analysis Intelligence Index provides a common quality measure; it also records how much time and money its own tasks consumed. Those measures should sit next to one another.
Claude Fable 5.2 leads the captured index in its Max configuration with default fallback. Look at the effort label before matching an endpoint. A model name can cover several reasoning settings, and an API quote in dollars per million tokens describes something different from the evaluator’s observed cost per task.
| Configuration | Index | USD per benchmark task |
|---|---|---|
| 1. Claude Fable 5.2 Max, Default Fallback | 65 | $26.54 |
| 2. GPT-6.2 Astra Max | 62 | $4.00 |
| 3. Gemini 4 Carbon High | 60 | $5.48 |
| 4. Claude Opus 5.5 Max, Default Fallback | 59 | $10.04 |
| 5. GPT-6.2 Sol Max | 57 | $1.01 |
| 6. Claude Sonnet 5.5 Max, Default Fallback | 55 | $10.02 |
| 7. Gemini 4 Hydrogen Xhigh | 54 | $3.16 |
| 8. GPT-6.1 Terra Max | 51 | $0.96 |
| 9. Gemini 4 Argon High | 50 | $4.97 |
| 10. Muse Wonder Full Max | 50 | - |
| 11. Kimi K3.1 Pro | 46 | $3.68 |
| 12. Muse Wonder Max | 44 | $6.95 |
| 13. Qwen4 Max | 43 | $3.27 |
| 14. Claude Haiku 5.5 Max | 36 | $0.41 |
| 15. Muse Ember Max | 36 | $0.14 |
| 16. Grok 4.7 Xhigh | 35 | $7.77 |
| 17. Tencent Hy4 | 35 | $1.03 |
| 18. DeepSeek V4.1 Flash 1017 Max | 33 | $0.60 |
| 19. GLM-5.3 Max | 32 | $5.01 |
| 20. Inkling 2 Small | 31 | $1.18 |
| 21. MiMo-V2.6-Pro | 30 | $0.26 |
| 22. Ling 3.1 Flash 1019 | 30 | $3.19 |
| 23. MiniMax-M3.1-Flash | 29 | $0.49 |
| 24. Qwen3.8-Flash-Next | 29 | $0.57 |
| 25. Step 5 Preview | 29 | $2.35 |
| 29. Mistral Large 4 High | 26 | $2.06 |
| 31. Agnes 2.5 Pro Theta | 24 | $0.57 |
| 32. DeepSeek V4 Pro 0813 Max | 23 | $1.77 |
| 33. Nanbeige4.3-33B | 20 | - |
| 35. JT-4.1 Flash 236B A21B | 19 | - |
| 38. Beam 1.0 | 18 | $0.28 |
| 105. Astra for Law 6.2 Max | - | - |
| 115. Nemotron 3.5 Bolt E4B 4GB Reasoning | - | - |
| 173. Gemini 3.1 Pro Preview | - | - |
Other releases in the selection
Claude Fable 5, Claude Opus 4.6, Claude Fable 5.1, Claude Opus 4.7, Gemini 3.8 Flash, Muse Spark 1.3, Muse Spark 1.2, Muse Spark 1.1, Claude Opus 5, Gemini 3.7 Flash, Muse Spark, Kimi K3, Gemini 3 Pro Preview, GPT-5.6 Sol, Qwen3.8 Max. These entries are included through the recent-release or Arena selection and have no corresponding position in this captured index table.
Time to the first token measures the initial wait. Throughput measures how quickly the rest of the answer arrives. For an example of the difference, see the Claude Fable model card.
Arena peer comparison
Arena uses its own matchups and snapshot. Keep these scores separate from the Artificial Analysis table.
| Model | Arena placement | Reported score |
|---|---|---|
| Gemini 4 Carbon (High) | 1 | 1526 |
| Gemini 4 Argon (High) | 2 | 1522 |
| Muse Wonder (Max) | 3 | 1511 |
| Claude Opus 5.5 (High) | 4 | 1507 |
| Claude Fable 5 (High) | 5 | 1504 |
| Claude Opus 4.6 (High) | 6 | 1503 |
| Claude Fable 5.1 (Max) | 7 | 1502 |
| Claude Opus 4.7 (High) | 8 | 1501 |
| Gemini 3.8 Flash (High) | 9 | 1497 |
| Muse Spark 1.3 (Max) | 10 | 1497 |
| Claude Opus 4.6 | 11 | 1496 |
| Claude Opus 4.7 | 12 | 1494 |
| Muse Spark 1.2 (xHigh) | 13 | 1493 |
| Muse Spark 1.1 | 14 | 1491 |
| Gemini 4 Hydrogen (High) | 15 | 1490 |
| Claude Opus 5 (High) | 16 | 1490 |
| Gemini 3.7 Flash (High) | 17 | 1489 |
| Muse Spark | 18 | 1489 |
| Claude Opus 5 (Max) | 19 | 1489 |
| Kimi K3 (Max) | 20 | 1487 |
| Gemini 3.1 Pro Preview | 21 | 1487 |
| Gemini 3 Pro | 22 | 1486 |
| GPT-5.6 Sol (xHigh) | 23 | 1485 |
| Kimi K3.1 Pro (Max) | 24 | 1485 |
| Qwen3.8-Max | 25 | 1484 |
Which is the most powerful LLM model? For the Artificial Analysis Intelligence Index snapshot, the highest-scoring configuration is Claude Fable 5.2 Max with default fallback. Change the task, available time or inference budget and the buying decision can change.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →Sources
- https://artificialanalysis.ai/leaderboards/models (accessed November 17, 2026)
- https://arena.ai/leaderboard/chat/text (accessed November 17, 2026)
- https://artificialanalysis.ai/models/claude-fable-5-2 (accessed November 17, 2026)
More on this from QUANTSLANT
benchmarks·Inworld models, access and benchmark guideInworld model and access guide with a practical US-English evaluation brief and links to publisher materials.