FrontierCS 2 Mean Score
Code · 2026-10-20
Mean of the task evaluators' bounded 0-100 final scores over the same 48 FrontierCS 2 runs, with a run that ends without a final submission scored 0. Each task scores on its own scale around the human reference, so this view is kept as a diagnostic and never enters capability.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 56.4 |
| GPT-6 Astra | 54.1 |
| GPT-6.1 Sol | 52.8 |
| Muse Wonder | 51.2 |
| Gemini 4 Argon | 49.1 |
| GLM-5.3 | 48.9 |
| Kimi K3 | 45.9 |
| Muse Spark 1.3 | 43.8 |
| GLM-5.3 Flash | 38.6 |
| Step 5 Preview | 36.4 |