APEX-Agents leaderboard
- Gemini 4 Argon leads
- 82.4%
- models measured
- 51
- lead over #2
- +6.9 pts
- median
- 49.2%
Leader on APEX-Agents · read October 22
On October 22, 2026, 51 of our catalog's models had a APEX-Agents result, and Google Gemini 4 Argon led with 82.4%, ahead of Anthropic Claude Sonnet 5.5 at 75.5%.
Epoch AI collects its results; its benchmarking hub describes the tasks and links the authors.
Role in the index
APEX-Agents is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Mean score | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|
| 1 | Gemini 4 ArgonGoogle | 82.4% | 87.4% | 81.0 | $10.00 | |
| 2 | Claude Sonnet 5.5Anthropic | 75.5% | 83.7% | 87.3 | $10.00 | |
| 3 | Claude Opus 5.5Anthropic | 73.5% | 81.3% | 88.8 | $20.00 | |
| 4 | Claude Fable 5.1Anthropic | 68.6% | 77.1% | 84.2 | $50.00 | |
| 5 | Gemini 3.7 FlashGoogle | 67.8% | 79.4% | 58.0 | $3.75 | |
| 6 | Claude Opus 5Anthropic | 65.8% | 77.0% | 83.0 | $25.00 | |
| 7 | Grok 4.6xAI | 65.3% | 77.3% | 65.7 | $6.00 | |
| 8 | GPT-6 AstraOpenAI | 64.7% | 75.1% | 85.1 | $50.00 | |
| 9 | Gemini 3.8 FlashGoogle | 64.3% | 72.9% | 61.6 | $3.75 | |
| 10 | Claude Fable 5Anthropic | 63.6% | 74.5% | 79.3 | $50.00 | |
| 11 | Qwen3.8 Max (0902)Qwen | 63.3% | 74.3% | 62.4 | $6.00 | |
| 12 | GPT-6.1 SolOpenAI | 60.0% | 73.2% | 82.4 | $10.00 | |
| 13 | Muse Spark 1.3Meta | 58.6% | 70.4% | 57.4 | $4.25 | |
| 14 | GPT-5.6 TerraOpenAI | 58.2% | 71.8% | 63.9 | $12.00 | |
| 15 | GLM 5.3Z.ai | 56.6% | 66.2% | 70.6 | $4.40 | |
| 16 | Grok 4.5xAI | 56.2% | 70.5% | 58.6 | $6.00 | |
| 17 | Claude Haiku 5.5Anthropic | 56.1% | 68.3% | 70.7 | $0.50 | |
| 18 | GPT-5.5OpenAI | 55.1% | 70.3% | 71.3 | $30.00 | |
| 19 | Grok 4.7xAI | 54.6% | 68.7% | 66.7 | $6.00 | |
| 20 | Claude Sonnet 5Anthropic | 54.5% | 67.4% | 57.7 | $10.00 | |
| 21 | MiMo-V2.6-ProXiaomi | 54.4% | 65.3% | 65.4 | $0.87 | |
| 22 | GPT-6 SolOpenAI | 54.3% | 68.7% | 76.0 | $10.00 | |
| 23 | GLM 5.3 FlashZ.ai | 52.8% | 66.8% | 57.7 | $0.50 | |
| 24 | GPT-5.4OpenAI | 52.4% | 68.2% | 64.4 | $15.00 | |
| 25 | GPT-5.6 Sol ProOpenAI | 51.4% | 65.2% | Not benchmarked | $20.00 | |
| 26 | Claude Opus 4.7Anthropic | 49.2% | 63.1% | 67.5 | $25.00 | |
| 27 | Claude Opus 4.8Anthropic | 48.9% | 64.4% | 65.0 | $25.00 | |
| 28 | Kimi K3Moonshot AI | 47.8% | 62.2% | 62.3 | $15.00 | |
| 29 | Qwen3.8 27BQwen | 47.5% | 62.2% | 74.2 | $2.55 | |
| 30 | DeepSeek V4 Pro 0813DeepSeek | 47.3% | 62.3% | 56.1 | $1.98 | |
| 31 | Gemini 3.6 FlashGoogle | 46.9% | 63.4% | 62.5 | $3.75 | |
| 32 | Claude Opus 4.6Anthropic | 46.3% | 61.8% | 57.2 | $25.00 | |
| 33 | Muse Spark 1.2Meta | 45.4% | 59.8% | 55.7 | $4.25 | |
| 34 | GLM 5.2Z.ai | 45.2% | 61.1% | 61.1 | $4.40 | |
| 35 | GPT-6 LunaOpenAI | 44.3% | 59.0% | 57.3 | $0.50 | |
| 36 | Claude Sonnet 4.6Anthropic | 43.0% | 57.6% | 57.3 | $15.00 | |
| 37 | GPT-5.6 LunaOpenAI | 43.0% | 57.8% | 58.0 | $1.20 | |
| 38 | GLM 5.1Z.ai | 40.9% | 56.8% | 49.2 | $4.40 | |
| 39 | DeepSeek V4.1 FlashDeepSeek | 39.5% | 50.5% | 62.5 | $0.84 | |
| 40 | MiniMax M3MiniMax | 37.7% | 53.5% | 39.2 | $1.20 | |
| 41 | Kimi K2.7 CodeMoonshot AI | 37.6% | 53.4% | 53.4 | $4.00 | |
| 42 | Gemini 3.1 Pro PreviewGoogle | 35.1% | 52.2% | 32.3 | $12.00 | |
| 43 | T InklingThinking Machines | 33.8% | 48.8% | 31.5 | $4.05 | |
| 44 | Muse Spark OpenMeta | 33.4% | 48.7% | 47.4 | $4.25 | |
| 45 | Muse Spark 1.1Meta | 31.8% | 46.2% | 70.3 | $4.25 | |
| 46 | Gemini 3.5 Flash LiteGoogle | 29.3% | 46.1% | 49.0 | $2.50 | |
| 47 | Gemini 3.5 FlashGoogle | 27.5% | 44.3% | 39.8 | $9.00 | |
| 48 | Qwen3.5 397B A17BQwen | 24.9% | 40.3% | Not benchmarked | $2.34 | |
| 49 | Nemotron 3 UltraNVIDIA | 22.7% | 38.3% | 30.0 | $2.40 | |
| 50 | DeepSeek V3.2DeepSeek | 21.3% | 35.1% | 30.0 | $0.88 | |
| 51 | gpt-oss-120bOpenAI | 4.4% | 8.3% | Not benchmarked | $0.50 |
Only models in our catalog are listed, and each appears once with its best published run. 215 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Every other column is a figure the board publishes for the same run.