List your company

Scientific Reasoning on Sydney Biology per-architecture breakdown (full)

8.93BLEURLearner-LLM
1.95163.76335.5757.3867May 6, 2026

Evaluation results

MethodLinks
RLearner-LLMBackbone=Gemma4-E4B2026.058.9388.0384.7254.2123.093.01424.667
Qwen3-8B SFTBackbone=Qwen3-8B2026.058.4487.8884.1659.2917.372.26—
Gemma4-E4B-it SFTBackbone=Gemma4-E4B-it2026.058.4487.9484.5262.0924.693.03645.804
RLearner-LLMBackbone=Qwen32026.054.1883.5379.9582.8322.842.78—
DPO v2Backbone=LLaMA-2-13B, Note=ours2026.053.6483.6784.2662.927.742.94746.37
RLearner-LLMBackbone=LLaMA-22026.053.1683.5986.263.2735.622.83764.06
ILearner-LLMK=52026.052.74—78.8964.518.373.1843—
SFTBackbone=LLaMA-2-13B2026.052.2282.4478.762.495.373.193719.001