IFM K2-Horizon-375B Card - SWE-bench Pro (Strict, No Internet): leaderboard
SWE-bench Pro asks agents to resolve real issues from professional software repositories. IFM's K2-Horizon-375B-A23B model card runs it in strict mode with no internet access, so values differ from standard SWE-bench Pro results. Resolved rate in percent, higher is better; effort settings are in the model labels.
Metric: Resolved (%). Source: huggingface.co. Saturation forecast: Around August 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Luna (Max) | 48.8 | #79 (GPT-5.6 Luna) |
| 2 | GLM-5.2 (Max) | 46.7 | #80 (GLM-5.2) |
| 3 | MiniMax-M3 | 43.8 | #151 |
| 4 | Inkling (xHigh) | 43.1 | #144 (Inkling) |
| 5 | Nemotron 3 Ultra | 38.7 | #245 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ifm-k2-horizon-375b-card-swe-bench-pro-strict-no-internet · How It Works · Data refreshed daily, snapshot 2026-11-05.