Tsinghua's ActionOPD lifts Qwen3-4B agent success on τ²-bench from 29.2% to 40%
jiqizhixin · x · 2026-11-17
Tsinghua University introduces ActionOPD in an arXiv preprint: a teacher-distillation scheme for agents that scores teacher actions against student candidates at states the student actually visits — requiring no teacher token probabilities and no tokenizer matching. Tool calls are compared by operation and normalized arguments; user-facing replies are checked for intent and key facts; the best candidate advances the environment and the candidate group supplies the policy update. With a Qwen3-32B teacher, Qwen3-4B's average single-trial success on held-out τ²-bench tasks rises from 29.2% to 40.0% in matched controlled comparisons (trained variants at step 50), though offline distillation still retains higher performance.
More from coding & agent
- Courses are becoming SaaS: packaging your method as agent skills buyers can run — VibeMarketer_ · 2026-11-17
- System One decision models take off: Jev creator rallies devs, Gemini Decisions API dissected — rseroter · 2026-11-17
- Resin: Personal Project Lets an AI Agent Write the Forward Pass and Out Pops a Standalone LLM Inference Engine — VoidAsuka · 2026-11-17
- Early hands-on with Gemini 4 Carbon in Antigravity CLI: slow but insightful — iskander · 2026-11-17
- Clinical genomics agent spots two lethal translocations NGS missed, in first peer-reviewed paper — danielmckinn0n · 2026-11-17
- Anthropic's daily code merges grew 12x in 4 months, now bottlenecked by human-scale infra — Dr_Singularity · 2026-11-17