Agentic coding
Resolving real repository and terminal tasks through a tool-using loop. Every score here measures a model together with a scaffold, so entries carry the scaffold as a variant and two rows for one model are expected rather than a duplicate.
Ordered by how many of these benchmarks place a model in their top 10, then by its best placing. Each column keeps its own order; nothing here is averaged.
| Model | In top 10 | Artificial Analysis LLM Leaderboard Terminal-Bench 2.1 | Artificial Analysis LLM Leaderboard Agentic Index | Artificial Analysis Coding Agents Artificial Analysis Coding Agent Index | Artificial Analysis Coding Agents Terminal-Bench v2 |
|---|---|---|---|---|---|
 GPT-5.6 Sol openai | 4/4 | #1 89.5% | #5 57.8 | #2 0.666 | #1 87.7% |
 Claude Opus 5 anthropic | 4/4 | #2 89.1% | #1 59.2 | #1 0.667 | #5 84.9% |
| GPT-5.6 Terra openai | 4/4 | #4 88.0% | #10 50.2 | #4 0.623 | #7 84.1% |
 Kimi K3 kimi | 4/4 | #6 85.0% | #8 54.3 | #6 0.613 | #8 83.7% |
 Grok 4.5 xai | 3/4 | #10 81.6% | #14 48.9 | #3 0.644 | #3 85.3% |
 Gemini 3.7 Flash google | 2/4 | #5 85.8% | #18 45.1 | #11 0.571 | #2 86.1% |
 GLM-5.3 zai | 2/4 | #8 83.9% | #2 59.1 | — | — |
| Grok 4.6 xai | 2/4 | #3 88.4% | #3 58.7 | — | — |
 Qwen3.8 Max alibaba | 2/4 | #11 81.3% | #4 58.4 | #9 0.598 | #11 79.4% |
| GPT-5.5 openai | 2/4 | — | — | #5 0.615 | #6 84.1% |
| Qwen3.8 2.4T A95B alibaba | 2/4 | #9 82.0% | #6 57.1 | — | — |
| Claude Fable 5 (Opus 4.8 fallback) anthropic | 2/4 | #7 84.6% | #7 56.6 | — | — |
| Claude Opus 4.8 anthropic | 2/4 | — | — | #7 0.606 | #9 81.3% |
| GPT-5.6 Luna openai | 2/4 | #12 80.9% | #16 46.9 | #10 0.587 | #10 79.8% |
 DeepSeek V4 Flash deepseek | 1/4 | — | — | #12 0.555 | #4 84.9% |
 Muse Spark 1.2 meta | 1/4 | #14 80.1% | #13 49.3 | #8 0.605 | #12 78.6% |
| Qwen3.8 27B alibaba | 1/4 | #15 79.8% | #9 50.9 | — | — |
A dash means the source does not cover that model, not that it scored zero. Models with no registry entry are excluded here because they cannot be joined across sources — they are still shown, flagged, on their own source's board.