OJBench (Online Judge Bench) — Competitive-programming judge benchmark, C++ subset; distinct from unscoped/python.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Gemini 3 Pro | 68.5 | 1 | 2026-06-15 | |
| 2 | Kimi K2.5 | Moonshot | 57.4 | 1 | 2026-06-15 |
| 3 | GPT-5 | OpenAI | 56.2 | 1 | 2026-05-15 |
| 4 | DeepSeek-V3.2 | DeepSeek | 54.7 | 2 | 2026-06-15 |
| 5 | Claude Opus 4.5 | Anthropic | 54.6 | 1 | 2026-06-15 |
| 6 | Kimi K2 (Reasoning) | Moonshot | 48.7 | 2 | 2026-05-15 |
| 7 | Claude Sonnet 4.5 | Anthropic | 30.4 | 1 | 2026-05-15 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.