Epoch AI - Frontierswe: leaderboard
Epoch’s mirror of FrontierSWE V2 mean@5 results under the Proximus harness and a stated maximum-effort policy. Exact API effort settings remain unverified for some primary labels. Fable headline values exclude the MSMS fallback contribution shown in the task matrix; preserve the official headline rather than recomputing from that matrix.
Metric: Score. Source: epoch.ai. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gpt-6-astra_max | 65.51 | |
| 2 | claude-opus-5-5_max | 62.33 | |
| 3 | claude-sonnet-5-5_max | 61.89 | |
| 4 | claude-fable-5-1_max | 56.29 | |
| 5 | gemini-4-argon_high | 54.96 | |
| 6 | claude-opus-5_max | 52.01 | |
| 7 | claude-fable-5_max | 46.96 | |
| 8 | claude-haiku-5-5_max | 43.77 | |
| 9 | gpt-5.6-sol_max | 32.2 | |
| 10 | glm-5.3_max | 30.18 |
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-frontierswe · How It Works · Data refreshed daily, snapshot 2026-10-20.