Program Bench / ProgramBench — Programming eval in GLM-5.2/Kimi-K2 cards; folds spelling variant. Distinct from PostTrainBench.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Kimi K3 | Moonshot | 77.8 | 3 | 2026-08-23 |
| 2 | GPT-5.6 Sol | OpenAI | 77.6 | 1 | 2026-07-27 |
| 3 | Claude Fable 5 | Anthropic | 76.8 | 1 | 2026-07-27 |
| 4 | Claude Opus 4.8 | Anthropic | 71.9 | 4 | 2026-07-27 |
| 5 | GPT-5.5 | OpenAI | 70.8 | 4 | 2026-07-27 |
| 6 | GLM-5.2 Full Open Source | Z.ai | 63.7 | 5 | 2026-08-23 |
| 7 | Kimi K2.7 Code | Moonshot | 53.6 | 2 | 2026-08-23 |
| 8 | GLM-5.1 | Z.ai | 50.9 | 2 | 2026-07-13 |
| 9 | DeepSeek-V4-Pro | DeepSeek | 47.8 | 2 | 2026-07-13 |
| 10 | Gemini 3.1 Pro | 39.5 | 2 | 2026-07-13 | |
| 11 | GLM-5.3 | Z.ai | 19 | 1 | 2026-08-23 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.