List your company
ResearchFormerly WizWand

Science Question Answering on FrontierScience

70.5AccuracyQwen3.5-397B
4.9821.993956.01Mar 22, 2026Apr 1, 2026Apr 11, 2026Apr 21, 2026May 1, 2026May 11, 2026May 22, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen3.5-397BBase Model=Qwen3.5-397B2026.0570.5
Qwen3.5-122BBase Model=Qwen3.5-122B2026.0559.8
Kimi-K2.5Base Model=Kimi-K2.52026.0555.2
Qwen3.5-35BBase Model=Qwen3.5-35B2026.0554.3
Deepseek-V3.2Base Model=Deepseek-V3.22026.0554.2
Qwen3.5-9B + MaRBase Model=Qwen3.5-9B, Optimization Method=MaR2026.0550
Qwen3.5-9BBase Model=Qwen3.5-9B2026.0542.4
Qwen3.5-9B + DAPOBase Model=Qwen3.5-9B, Optimization Method=DAPO2026.0539
GLM-5.1Base Model=GLM-5.12026.0538.2
ARYAPrompting Strategy=Zero-shot prompting, Parameters=02026.0337.5
Qwen3.5-4BBase Model=Qwen3.5-4B2026.0534.7
GPT-OSS-120BBase Model=GPT-OSS-120B2026.0534.4
Qwen3.5-4B + MaRBase Model=Qwen3.5-4B, Optimization Method=MaR2026.0534
Qwen3.5-4B + DAPOBase Model=Qwen3.5-4B, Optimization Method=DAPO2026.0533
GPT-5.2Prompting Strategy=Optimized prompting / few-shot / CoT2026.0325.8
GPT-5.2 (pub)Context=Best Published Baseline2026.0325.8
Claude Opus 4.6Prompting Strategy=Zero-shot prompting2026.038.8
GPT-5.2Prompting Strategy=Zero-shot prompting2026.037.5