30B Open Model OpenResearcher Beats GPT-4.1 on BrowseComp-Plus
TheZachMueller · x · 2026-09-11
Lambda announced that OpenResearcher, a 30B open model trained entirely on offline research trajectories, outperformed GPT-4.1, Claude-4-Opus and DeepSeek-R1 on BrowseComp-Plus.
OpenResearcher scored 54.8%, a 34-point gain over its base model, and transferred to the live web without any live-web training data.
More from Models
- NVIDIA ships NVFP4-quantized Qwen3.8-27B, trending on Hugging Face — nvidia · 2026-09-11
- Meta's Muse Spark 1.3 coding model lands in Cursor, claims Pareto frontier on CursorBench — parth007_96 · 2026-09-11
- Reddit User: Astra-6 Disappoints at Web Design While Claude Fable Shines — pivo161 · 2026-09-11
- AI Writes Entire 3D Game Engine Overnight in Bend2, Hitting 120 FPS — rickasaurus · 2026-09-11
- Eval lab says Anthropic's top model cheats ~5x more than rival Astra — steipete · 2026-09-11
- GPT-6 Astra users report aggressive token burn compared to GPT-5.6 Sol — gillu-21 · 2026-09-11