SearchJev: a fast System-1 model for search agents with 5.3x faster decisions and 45%→54% accuracy
CShorten30 · x · 2026-10-18
The team unveils SearchJev, a fast and calibrated System-1 model for search agents, released as open source.
Key numbers:
- 5.2–5.3x faster decisions
- 41–74% lower calibration error
- BrowseComp-Plus accuracy improved from 45% to 54%
The idea is to replace slow multi-step reasoning in search agents with calibrated System-1 style fast decisions while preserving accuracy.
More from coding & agent
- Clicky MCP Server queries web analytics via natural language — modelcontextprotocol · 2026-10-18
- CipherHUB ships 31 MCP cryptography tools spanning PQC and KMS — modelcontextprotocol · 2026-10-18
- Codex's new tab completion predicts your next move, and users love it — mark_k · 2026-10-18
- Open-source voxel editor ships MCP server letting Claude build and self-check 3D models — AbaGuy17 · 2026-10-18
- Dev ships AI-written game to 3 portals in an afternoon, Opus 5.5 did all the coding — Roger_M_Taylor · 2026-10-18
- 10 open-source repos to build a one-person AI content factory, from scripts to voiceovers — Roger_M_Taylor · 2026-10-18