FrontierSWE: leaderboard
Ultra-long-horizon coding agent benchmark (20h per task) testing implementation, performance engineering, and ML research at the edge of human ability. 17 tasks scored on a continuous 0-1 scale with 5 trials each.
Metric: Dominance (%). Source: www.frontierswe.com. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 88 | #16 |
| 2 | GLM-5.3 | 78 | #39 |
| 3 | Grok 4.6 | 78 | #26 |
| 4 | Grok 4.5 | 72 | #48 |
| 5 | Claude Opus 4.8 | 67 | #37 |
| 6 | GLM-5.2 | 67 | #73 |
| 7 | GPT-5.5 | 65 | #31 |
| 8 | Claude Opus 4.7 | 56 | #49 |
| 9 | Claude Opus 4.6 | 49 | #63 |
| 10 | GPT-5.4 | 46 | #79 |
| 11 | Gemini 3.1 Pro (Preview) | 34 | #58 |
| 12 | Composer 2.5 | 34 | #122 |
| 13 | GLM-5.1 | 26 | #97 |
| 14 | DeepSeek V4 Pro | 25 | #102 |
| 15 | Kimi K2.5 | 23 | #144 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=frontierswe · How It Works · Data refreshed daily, snapshot 2026-10-20.