Benchmarks · coding
FrontierSWE v1
Original FrontierSWE leaderboard (17 tasks): dominance = win rate against a random opponent on a random task, mean@5, each model in its native CLI harness. Preserved as published.
17 models · 17 results · data as of 2026-10-29 · frontierswe.com ↗
Leaderboard
Fable 5 max
All results
17 rows| # | Model | Effort | FrontierSWE v1 | Harness |
|---|---|---|---|---|
| 1 | Fable 5 | max | 88.2% | Claude Code |
| 2 | GLM 5.3 | max | 78.1% | Claude Code |
| 3 | Grok 4.6 | max | 77.9% | Grok CLI |
| 4 | Grok 4.5 | max | 72.1% | Grok CLI |
| 5 | GLM 5.2 | max | 67.5% | Claude Code |
| 6 | Opus 4.8 | max | 66.5% | Claude Code |
| 7 | GPT-5.5 | xhigh | 64.5% | Codex |
| 8 | Opus 4.7 | max | 56.3% | Claude Code |
| 9 | Opus 4.6 | max | 48.9% | Claude Code |
| 10 | GPT-5.4 | xhigh | 46.0% | Codex |
| 11 | Gemini 3.1 Pro | high | 34.4% | Gemini CLI |
| 12 | Composer 2.5 | none | 33.8% | Cursor CLI |
| 13 | GLM 5.1 | max | 25.7% | Claude Code |
| 14 | DeepSeek V4 Pro | max | 24.6% | Claude Code |
| 15 | Kimi K2.5 | max | 23.3% | Kimi CLI |
| 16 | Kimi K2.6 | max | 22.2% | Kimi CLI |
| 17 | Qwen 3.6 Plus | max | 19.9% | Qwen Code |