MERA Code - UnitTests: leaderboard
MERA Code (Russian) UnitTests task: writing unit tests that actually exercise the supplied behaviour. Each task keeps its own metric — pass rates, CodeBLEU and chrF all appear across this suite — so read the metric label before comparing boards.
Metric: CodeBLEU (%). Source: mera.a-ai.ru. Saturation forecast: Around 2033. 66 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 34.21 | #144 |
| 2 | GLM-5.1 | 33.23 | #97 |
| 3 | GPT-5.1 Codex Max | 32.31 | #100 |
| 4 | GPT-5.4 | 32.1 | #79 |
| 5 | DeepSeek V4 Flash | 31.95 | #137 |
| 6 | DeepSeek V4 Pro | 31.8 | #102 |
| 7 | Qwen 3.6 35B A3B | 30.46 | #226 |
| 8 | Qwen 3 32B | 30.28 | #443 |
| 9 | Qwen 3 14B | 30 | #537 |
| 10 | Claude Opus 4.6 | 29.92 | #63 |
| 11 | Qwen 3 30B A3B 2507 (Thinking) | 28.75 | #376 (Qwen 3 30B A3B 2507) |
| 12 | Qwen 3.6 27B | 28.31 | #170 |
| 13 | Qwen 3.5 2B | 26.57 | #1148 |
| 14 | Qwen 3 8B | 26.52 | #674 |
| 15 | Qwen 3 Next 80B A3B (Thinking) | 26.51 | #320 (Qwen 3 Next 80B A3B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mera-code-unittests · How It Works · Data refreshed daily, snapshot 2026-10-20.