Write CLI, scripts, and DevOps
Pick the best model for scripts and DevOps from terminal agent tasks, general long-horizon engineering, and mergeability.
Right now Claude Opus 5.5 is first on 1 of 2 boards it enters.
Sorted by compositeSynced 2026-10-10
| Claude Opus 5.5 | 96 | 2/3 | 1 | — | 2 |
| GPT-6 Astra | 93 | 3/3 | 3 | 1 | 3 |
| Claude Sonnet 5.5 | 91 | 2/3 | 2 | — | 4 |
| GPT-6.1 Sol | 87 | 2/3 | 3 | — | 5 |
| Claude Fable 5.1 | 82 | 2/3 | 5 | — | 7 |
| Claude Opus 5 | 80 | 3/3 | 11 | 3 | 11 |
| GPT 6 Sol | 76 | 2/3 | 16 | — | 6 |
| Claude Fable 5 | 75 | 2/3 | 18 | 5 | — |
| GPT 5.6 Sol | 69 | 3/3 | 22 | 4 | 10 |
| GLM-5.3 | 65 | 3/3 | 20 | 7 | 18 |
| Grok 4.7 | 63 | 2/3 | 21 | — | 8 |
| GLM-5.3-Flash | 61 | 1/3 | 23 | — | — |
| Gemini 3.8 Flash | 56 | 2/3 | 29 | 2 | — |
| GPT 5.6 Terra | 56 | 2/3 | 27 | 6 | — |
| Qwen3.8 Max | 55 | 2/3 | 25 | 16 | — |
| GPT 5.6 Luna | 52 | 2/3 | 30 | 10 | — |
| Opus 4.8 | 49 | 1/3 | 26 | — | — |
| Grok 4.6 | 46 | 1/3 | 28 | — | — |
| Claude Sonnet 5 | 42 | 2/3 | 33 | 18 | — |
| GPT 6 Luna | 41 | 2/3 | 31 | — | 16 |
Suggested build
Why it matters
First is not the cheapest
Opus 5.5 passes 64.8% at $4.7k a run. GPT-6.1 Sol ties for 3rd at 58.2% and $634.24.
Second costs more
Sonnet 5.5 passes 61.8% at $7.3k a run, more than first place.
This board ignores the interface
Whether a screen looks right is LMArena WebDev. This board only asks if the terminal task gets finished.