Baseten details deep research agent training: breadth-rewarding RL plus SFT for report writing
baseten · x · 2026-11-11
Baseten published a technical blog explaining how it built its deep research agent by splitting the problem into two subproblems:
- Research: trained with RL against a reward that credits breadth — distinct searches and sources — rather than just the final result.
- Report writing: taught mainly via SFT on strong reports built from the model's own evidence, followed by an end-to-end RL pass.
The divide-and-conquer recipe — breadth-rewarding RL for search, self-generated-data SFT plus end-to-end RL for writing — offers a concrete template for training deep research agents. Details in the full blog post.
More from coding & agent
- Ethan Mollick lets a swarm of agents argue their way into building a fantasy transport sim — emollick · 2026-11-11
- Margin: open-source Mintlify-compatible docs framework ships free hosting built for AI agents — ycombinator · 2026-11-11
- Netlify launches Servers: long-lived WebSocket servers in the same deploy as your site — thisiskp_ · 2026-11-11
- Duke and Princeton paper: 10 small parallel agent teams beat best-of-N and one giant team — rohanpaul_ai · 2026-11-11
- ERP team open-sources Dory docs-with-MCP and UX ShadCN theme library for free — hkdeman · 2026-11-11
- LangChain's OpenWiki builds citation-based evidence graphs for agent memory — a leaner GraphRAG alternative — JeremyCMorgan · 2026-11-11