KOCO-BENCH: leaderboard
KOCO-BENCH evaluates LLMs on domain-specialized software development using text-based code generation and knowledge-understanding tasks. The benchmark provides curated domain knowledge corpora alongside function- to project-level coding problems. Its pinned metric is Average Pass@1 (%), measuring first-attempt success on the evaluated tasks in the reported result table. Source: BenchmarkList’s attributed result table.
Metric: Average Pass@1 (%). Source: benchmarklist.com. Saturation forecast: Around 2035. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2 | 8.9 | #252 |
| 2 | Gemini 2.5 Pro | 8.5 | #159 |
| 3 | Qwen 2.5 Coder 7B Instruct | 7.5 | #958 |
| 4 | O4 Mini | 7.1 | #185 |
| 5 | GPT-5 Mini | 6.6 | #193 |
| 6 | Claude Sonnet 4.5 | 6.1 | #148 |
| 7 | DeepSeek V3.1 | 6.1 | #275 |
| 8 | Qwen 2.5 Coder 32B Instruct | 5.2 | #534 |
| 9 | Llama 3.1 8B Instruct | 5.1 | #1017 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=koco-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.