# Kagi LLM Benchmarking Project

Kagi, a search company, regularly quizzes AI models on a private set of tricky reasoning, coding and language questions that change over time, to see which ones actually think well.

Category: aggregators. Type: leaderboard. Status: active.
Metric: Share of benchmark tasks answered correctly (%) (higher is better).
Source: Kagi Search — https://help.kagi.com/kagi/ai/llm-benchmark.html, retrieved 2026-11-17, license: not stated.

| Rank | Model | Accuracy (%) | Cost ($, whole suite) | Provider |
|---|---|---|---|---|
| 1 | claude-5-fable-thinking | 91.4 | 10.4 | kagi (ult) |
| 2 | claude-5-fable | 88.8 | 9.9 | kagi (ult) |
| 3 | claude-4-8-opus-thinking | 88.8 | 8.8 | kagi (ult) |
| 4 | gpt-5-5 | 88.8 | 2.2 | kagi (ult) |
| 5 | claude-4-6-opus-thinking | 83.6 | 4.8 | kagi (ult) |
| 6 | grok-4-5 | 83.5 | 1.0 | kagi (ult) |
| 7 | claude-4-7-opus-thinking | 80.7 | 2.8 | kagi (ult) |
| 8 | claude-4-5-opus-thinking | 80.2 | 4.5 | kagi (ult) |
| 9 | gemini-3-pro | 80.1 | 0.4 | kagi (ult) |
| 10 | kimi-k2-5-thinking | 78.5 | 0.5 | kagi |
| 11 | gpt-5-pro | 76.8 | 31.8 | kagi (ult) |
| 12 | glm-5-thinking | 75.0 | 2.0 | openrouter |
| 13 | grok-4-20-thinking | 75.0 | 1.3 | kagi (ult) |
| 14 | chatgpt-4o (points to gpt 5.3) | 75.0 | 0.8 | kagi (ult) |
| 15 | claude-4-opus-thinking | 74.3 | 22.4 | kagi (ult) |
| 16 | qwen-3-5-397b | 73.7 | 1.8 | together |
| 17 | grok-4 | 73.6 | 1.0 | kagi (ult) |
| 18 | gpt-5-2 | 73.3 | 1.4 | kagi (ult) |
| 19 | claude-4-7-opus | 73.3 | 2.2 | kagi (ult) |
| 20 | claude-4-sonnet-thinking | 73.0 | 5.4 | kagi (ult) |
| 21 | gpt-5 | 72.7 | 7.1 | kagi (ult) |
| 22 | qwen3-max | 72.5 | 1.9 | openrouter |
| 23 | claude-4-6-opus | 72.4 | 1.6 | kagi (ult) |
| 24 | o3-pro | 72.1 | 34.2 | kagi (ult) |
| 25 | claude-4-5-opus | 70.7 | 1.4 | kagi (ult) |

Notes: Reasoning benchmark table, 116 tasks, last task list revision Feb 13, 2026; page has no explicit update date but lists models released after that. Order as published (by accuracy).
