CMASS: New agentic search benchmark with 423 QA pairs over 550M docs
lintool · x · 2026-10-29
Following BrowseComp-Plus, researchers released CMASS, a new benchmark for agentic search. It has 423 question–answer pairs drawn from BrowseComp-Plus, DeepSearchQA, Total Recall QA and HotpotQA, all projected onto ClimbMix—a realistic corpus of over 550M documents. Search agents must therefore operate over a web-scale, realistic corpus, making this a tougher test for agentic retrieval systems.
More from coding & agent
- Microsoft launches Copilot Fusion: frontier-level quality, 40% faster, up to 77% cheaper — satyanadella · 2026-10-29
- If an agent could just click around and do it, what's left of your startup? — dunkhippo33 · 2026-10-29
- Hustle Fund partner: vibe coding kills the $25k CRM, support is the real moat — dunkhippo33 · 2026-10-29
- GitHub launches Copilot Projects: project-level AI agent spanning multiple repos — 0xkarasy · 2026-10-29
- Lor enters private beta, aiming to replace an entire AI lab with hundreds of autonomous agents — vjkaruna · 2026-10-29
- Fizgig H3 still-image latent node drops into any ComfyUI workflow — and runs faster — robertwellesley · 2026-10-29