Code Generation on LiveCodeBench (solve rate)
48Solve RateDecompRL
Evaluation Results
| Method | Links | |
|---|---|---|
| DecompRLToken budget=500,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 48 | |
| lme16Token budget=500,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 46 | |
| pass@8Token budget=500,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 46 | |
| DecompRLToken budget=100,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 44 | |
| GRPOToken budget=500,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 44 | |
| lme48Token budget=500,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 44 | |
| pass@8Token budget=100,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 41 | |
| DecompRLToken budget=50,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 40 | |
| GRPOToken budget=100,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 40 | |
| lme16Token budget=100,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 40 | |
| pass@8Token budget=50,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 39 | |
| lme48Token budget=100,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 39 | |
| instructToken budget=500,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 39 | |
| GRPOToken budget=50,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 38 | |
| lme16Token budget=50,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 38 | |
| lme48Token budget=50,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 36 | |
| instructToken budget=100,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 33 | |
| GRPOToken budget=10,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 32 | |
| lme16Token budget=10,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 31 | |
| pass@8Token budget=10,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 31 | |
| lme48Token budget=10,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 30 | |
| instructToken budget=50,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 30 | |
| GRPOToken budget=5,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 29 | |
| lme16Token budget=5,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 28 | |
| pass@8Token budget=5,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 27 | |
| lme48Token budget=5,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 26 | |
| DecompRLToken budget=10,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 25 | |
| instructToken budget=10,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 21 | |
| lme16Token budget=1,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 19 | |
| GRPOToken budget=1,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 18 | |
| DecompRLToken budget=1,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 18 | |
| DecompRLToken budget=5,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 18 | |
| instructToken budget=5,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 16 | |
| pass@8Token budget=1,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 15 | |
| lme48Token budget=1,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 12 | |
| instructToken budget=1,000 tokens, Backbone=Qwen 2.5 7B2026.07 | 6 |