LiveCodeBench Hard
The most difficult competition-level programming evaluation set within LiveCodeBench. Its problems come from recent competition platforms and emphasize multi-step reasoning and code correctness.
Timeline (last 90 days)
GVS5H claims that five Qwen3.8-27B instances working in coordination match Fable 5 on LiveCodeBench Hard.
LiveCodeBench Hard-level problems typically require multi-step reasoning, complex algorithm design, and handling of edge cases, placing extremely high demands on the model's one-shot code correctness rate.
All Facts (2)
GVS5H claims that five Qwen3.8-27B instances working in coordination match Fable 5 on LiveCodeBench Hard.
50%UnverifiedLiveCodeBench Hard-level problems typically require multi-step reasoning, complex algorithm design, and handling of edge cases, placing extremely high demands on the model's one-shot code correctness rate.
50%