ResearchBenchmarksCapability Elicitation on Super GPQA (test)Follow99Elicitation RateSFT+RL27.2445.8764.583.13Apr 23, 2026Evaluation ResultsMethodMethodLinksElicitation RateSFT+RLSupervisor=LLAMA 3.1 8...Supervisor=LLAMA 3.1 8B (VERY WEAK)2026.0499SFT+RLSupervisor=GPT 4o mini...Supervisor=GPT 4o mini (WEAK)2026.0497SFTSupervisor=GPT 4o mini...Supervisor=GPT 4o mini (WEAK)2026.0491SFTSupervisor=LLAMA 3.1 8...Supervisor=LLAMA 3.1 8B (VERY WEAK)2026.0475RLSupervisor=GPT 4o mini...Supervisor=GPT 4o mini (WEAK)2026.0430RLSupervisor=LLAMA 3.1 8...Supervisor=LLAMA 3.1 8B (VERY WEAK)2026.0430