Artificial Analysis Coding Agent Index v1.5
Indexes
Artificial Analysis Coding Agent Index v1.5 scores coding-agent and model pairs, such as Claude Code or Codex with a given model and effort, as the equal-weight average of pass@1 averaged over three attempts per task on DeepSWE v1.1 (113 tasks), Terminal-Bench 4.0 (66 tasks) and SWE-Atlas-QnA (124 tasks). Scores belong to the agent harness and model together. It is separate from the model-level Artificial Analysis Coding Index.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Sonnet 5.5 | 68.4 |
| Muse Wonder Full | 68.0 |
| Claude Opus 5.5 | 66.0 |
| Gemini 4 Argon | 63.8 |
| Muse Wonder | 63.7 |
| GPT-6.1 Sol | 62.9 |
| Claude Fable 5.1 | 62.2 |
| GPT-6 Astra | 61.6 |
| Claude Opus 5 | 59.7 |
| GPT-6 Sol | 56.7 |
| Grok 4.7 | 56.3 |
| GPT-5.6 Sol | 54.6 |
| Muse Spark 1.3 | 54.3 |
| GLM-5.3 | 53.6 |
| Kimi K3 | 51.9 |
| Step 5 Preview | 51.2 |
| Grok 4.6 | 47.0 |
| GPT-5.6 Luna | 43.2 |
| DeepSeek V4 Pro 0813 | 43.1 |
| Gemini 3.8 Flash | 41.9 |
| Claude Haiku 5.5 | 41.4 |
| Muse Spark Open | 41.2 |
| GPT-6 Luna | 41.1 |
| DeepSeek-V4-Flash-0731 | 38.7 |
| Beam 1.0 501B A23B | 27.7 |