Weaviate adds per-request effort levels to scale test-time compute in search
CShorten30 · x · 2026-10-27
Weaviate's Query Agent Search Mode now has an effort parameter, letting developers pick medium, high, or ultrahigh compute per request to balance quality, latency, and cost — cheap for easy lookups, more work for hard research questions.
- The three tiers scale computation at two stages: how long the agent thinks about decomposing queries and how thoroughly it reranks.
- On BRIGHT Biology, a hard reasoning-intensive retrieval benchmark, ultrahigh effort lifts nDCG@10 to 57.5 versus 13.0 for Hybrid Search alone.
- Available in weaviate-agents 1.8.0 and agents-typescript-client 1.7.0 with a single parameter change.
More from coding & agent
- Superuser first look: Grok bot meets Animal Crossing and Minecraft in agent demo — zeeg · 2026-10-27
- GameBeast: 73% More Users Reach Analytics via Claude, Cursor Than Built-in Assistant — zanbrum · 2026-10-27
- Claude Duet workflow: auto-investigation, Notion board and Linear tickets, fixed in 15 minutes — sloppenheimer · 2026-10-27
- Running Qwen 27B locally via MCP + Docker sandbox to refactor shell scripts — ag789 · 2026-10-27
- 8 agents beat 1 by 2.3x in valid experiments per hour, six-hour local swarm benchmarks — cong_ml · 2026-10-27
- How the swarm collaborates: shared best program plus rotating instruction variety — cong_ml · 2026-10-27