BenchmarkLiveCodeBench v6

LiveCodeBench v6

A dynamic evaluation benchmark for code generation and reasoning capabilities that continuously updates programming problems to prevent data contamination.

Timeline (last 90 days)

Oct 30

Pruning-based weakening raises the accuracy of Qwen3.5-4B on LiveCodeBench v6 from 39.1% to 56.7%.

Unverified50%

All Facts (1)

Unverified

Pruning-based weakening raises the accuracy of Qwen3.5-4B on LiveCodeBench v6 from 39.1% to 56.7%.

50%

Source Articles

Unsloth Releases Quantized DeepSeek-V4.1 GGUF: Low-Bit Precision Beats ISTAOct 30 Reverse Self-Distillation: Weakening a Model Can Actually Improve Its ReasoningOct 30