LLM rankings
- models measured
- 213
- lead over #2
- +1.6
- median
- 18.5
Leader on AA Intelligence Index · read October 20
Coding is our flagship ranking: an index over four coding benchmarks, the evidence behind each score and what a solved task costs. The other 7 axes each rank models on one public board, shown as published, with its date.
Every ranking
Coding is our index over four coding benchmarks; every other ranking is one public board, shown as published. Each row opens its full ranking.
| Ranking | Leader | Score | Next | Ranked | Board updated |
|---|---|---|---|---|---|
| CodingVibe Coding Index, four coding benchmarks | 1 Claude Opus 5.5 | 92.2 | 2Claude Fable 5.189.13Claude Sonnet 5.588.9 | 136 | October 20 |
| IntelligenceAA Intelligence Index | 1 Claude Opus 5.5 | 57.6 | 2Claude Sonnet 5.556.03Claude Fable 5.153.4 | 213 | October 20 |
| Agentic workAA Agentic Index | 1 Claude Fable 5.1 | 57.9 | 2Claude Opus 556.53Qwen3.8 Max (0902)56.0 | 87 | October 20 |
| ReasoningGPQA Diamond | 1 Gemini 3.1 Pro Preview | 95.5% | 2GPT-5.6 Sol95.2%3Grok 4.694.7% | 91 | September 1 |
| MathAIME | 1 Gemini 3.1 Pro Preview | 98.1% | 2GPT-5.296.9%3GPT-5.496.7% | 54 | April 16 |
| KnowledgeMMLU Pro | 1 Claude Fable 5.1 | 92.4% | 2Claude Opus 591.6%3Claude Fable 591.5% | 92 | September 1 |
| Web designDesign Arena: websites | 1 Muse Wonder | 1360 | 2Muse Spark 1.313563GPT-6 Astra1338 | 131 | October 20 |
| GeneralVals Index | 1 Claude Sonnet 5.5 | 70.8% | 2Claude Opus 5.570.2%3Gemini 4 Argon69.2% | 20 | October 19 |
The makers' medal table
In every category the makers are placed by their best model: gold, silver, bronze. Equal scores share a place, and a maker with no model on a board says so.
Why coding is ranked differently
Coding has four public benchmarks that publish results for the same models, so we can average them into one index and show how many of the four stand behind each score. It also has the only board that publishes what an attempt costs, which is how we compute the cost of a solved task.
Reasoning, math, knowledge and general skill each have one board in our data. A blend of one is that board, so we show it as published, with its date and the limits its publisher states, and keep it out of the coding index. Methodology