FrontierSWE V2: leaderboard
The second FrontierSWE round, September 2026: 34 ultra-long-horizon tasks across implementation, performance engineering, scientific computing, visual reasoning and AI research, such as reimplementing git in Zig, decoding speech from MEG recordings, training a medium-range weather model. Each task gets a 20-hour budget and five runs, and the board reports the mean of those five rather than V1’s pairwise dominance.
Metric: Mean@5 Score (%). Source: www.frontierswe.com. Source: epoch.ai. Saturation forecast: Around June 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 65.51 | #6 (GPT-6 Astra) |
| 2 | Claude Opus 5.5 (Max) | 62.33 | #3 (Claude Opus 5.5) |
| 3 | Claude Sonnet 5.5 (Max) | 61.89 | #11 (Claude Sonnet 5.5) |
| 4 | Claude Fable 5.1 (Max) | 56.29 | #9 (Claude Fable 5.1) |
| 5 | Gemini 4 Argon (High) | 54.96 | #8 (Gemini 4 Argon) |
| 6 | Claude Opus 5 (Max) | 52.01 | #12 (Claude Opus 5) |
| 7 | Claude Fable 5 (Max) | 46.96 | #16 (Claude Fable 5) |
| 8 | Claude Haiku 5.5 (Max) | 43.77 | #40 (Claude Haiku 5.5) |
| 9 | GPT-5.6 Sol (Max) | 32.2 | #19 (GPT-5.6 Sol) |
| 10 | GLM-5.3 (Max) | 30.18 | #39 (GLM-5.3) |
| 11 | Grok 4.7 (xHigh) | 29.5 | #25 (Grok 4.7) |
| 12 | Kimi K3 (Max) | 25.87 | #30 (Kimi K3) |
| 13 | Muse Spark 1.3 (Max) | 25.8 | #24 (Muse Spark 1.3) |
| 14 | Grok 4.6 (xHigh) | 25.29 | #26 (Grok 4.6) |
| 15 | Gemini 3.7 Flash (High) | 20.26 | #23 (Gemini 3.7 Flash) |
No result here: #7 GPT-6.1 Sol, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=frontierswe-v2 · How It Works · Data refreshed daily, snapshot 2026-10-20.