Training SID-1 to beat GPT-5 at search with 1k+ QPS RL
SID-1 is an agentic search model that is 24x faster than GPT-5.1-high, 374x cheaper than Sonnet 4.5, and achieves 1.9x higher recall than traditional RAG pipelines. Here's how we trained it using large-scale RL on turbopuffer.
Training SID-1 to beat GPT-5 at search with 1k+ QPS RL NEW: Instant namespace branching NEW: Branching for instant, copy-on-write namespaces Training SID-1 to beat GPT-5 at search with 1k+ QPS RL May 20, 2026 • Max Rumpf (Co-founder of SID), Sam Dauncey (Researcher at SID) guest Given sufficient search tools and time, humans can find almost anything. We search, read results, adapt, and search again until we find the information we seek. We're Max and Sam, co-creators of SID-1 , an agentic search model that builds upon this idea. As a result of its training, SID-1 nearly doubles recall over cla
Explore this link on the map →saved by
related reading
- Chroma Context-1: Training a Self-Editing Search Agent | Chromatrychroma.com
- Stream of Search (SoS): Learning to Search in Languagearxiv.org
- DeepSeek-R1arxiv.org
- Composer2.pdfcursor.com
- Introducing SWE-grep and SWE-grep-mini: RL for Multi-Turn, Fast Context Retrieval | Cognitioncognition.ai
- Patterns for Building LLM-based Systems & Productseugeneyan.com
- AI Model & API Providers Analysis | Artificial Analysisartificialanalysis.ai
- State of RL for reasoning LLMs | A. Weersaweers.de
- castform - the training platform for the ai engineercgft.io
- DeepSeek R1's recipe to replicate o1 and the future of reasoning LMsinterconnects.ai
- Our AI Research: How We Evaluate Semantic Search Technology | Exa Blogexa.ai
- PostTrainBenchposttrainbench.com