List your company
ResearchFormerly WizWand

Coding on LiveCodeBench (Throughput/Speedup)

3,413ThroughputSpecBundle
60.768931.0591,801.352,671.641Mar 19, 2026
Updated 7mo ago

Evaluation results

MethodLinks
SpecBundleTarget Model=Qwen-30B-A3B, Draft Model=SpecBundle, #GPUs=42026.033,4132.29
SpecBundleTarget Model=Llama-4-Scout, Draft Model=SpecBundle, #GPUs=82026.032,170.24.48
Existing Speculative DecodingTarget Model=Llama-4-Scout, Draft Model=Existing, #GPUs=82026.031,601.33.31
AutoregressiveTarget Model=Qwen-30B-A3B, Draft Model=None, #GPUs=42026.031,492.61
SpecBundleTarget Model=Llama-3.3-70B, Draft Model=SpecBundle, #GPUs=42026.031,459.42.6
SpecBundleTarget Model=Ling-Flash-V2, Draft Model=SpecBundle, #GPUs=82026.031,366.41.77
Existing Speculative DecodingTarget Model=Llama-3.3-70B, Draft Model=Existing, #GPUs=42026.031,303.42.32
SpecBundleTarget Model=Qwen-235B-A22B, Draft Model=SpecBundle, #GPUs=82026.031,155.71.93
SpecBundleTarget Model=Kimi-K2, Draft Model=SpecBundle, #GPUs=82026.03904.41.81
Existing Speculative DecodingTarget Model=Qwen-235B-A22B, Draft Model=Existing, #GPUs=82026.03803.81.34
AutoregressiveTarget Model=Ling-Flash-V2, Draft Model=None, #GPUs=82026.03770.41
AutoregressiveTarget Model=Qwen-235B-A22B, Draft Model=None, #GPUs=82026.03598.21
AutoregressiveTarget Model=Llama-3.3-70B, Draft Model=None, #GPUs=42026.03560.91
SpecBundleTarget Model=Llama-3.1-8B, Draft Model=SpecBundle, #GPUs=12026.03516.92.72
AutoregressiveTarget Model=Kimi-K2, Draft Model=None, #GPUs=82026.03500.11
AutoregressiveTarget Model=Llama-4-Scout, Draft Model=None, #GPUs=82026.03484.31
Existing Speculative DecodingTarget Model=Llama-3.1-8B, Draft Model=Existing, #GPUs=12026.03398.42.1
AutoregressiveTarget Model=Llama-3.1-8B, Draft Model=None, #GPUs=12026.03189.71