Tencent Open-Sources R3-Skill Retrieval Stack
tomaarsen · x · 2026-07-14
Tencent released R3-Skill, a two-stage retrieval stack designed specifically for LLM agent skills routing. It targets skill files like Anthropic's SKILL.md rather than traditional RAG text retrieval.
It consists of two 0.6B models, both under the Apache 2.0 license:
- R3-Embedding-0.6B: A bi-encoder for recall, based on Qwen3-Embedding-0.6B
- R3-Rerank-0.6B: A cross-encoder for reranking, based on Qwen3-Reranker-0.6B
The author emphasizes that these models aren't meant for general text retrieval. They are fine-tuned specifically for skill retrieval—selecting the appropriate skills within an agent workflow.
More from coding & agent
- Soft Clamp cuts tool-call overuse in multi-teacher distillation, from 13.7% to 9.0% — antgroup · 2026-07-21
- Agent harness memory loss and compaction are still a major usability problem — adityaag · 2026-07-21
- SpecJudge runs locally on Ollama to pick the right-sized AI model for your project — jokiruiz · 2026-07-21
- A developer maps out six design rules for CLIs that humans and AI agents can both use — yujiezha · 2026-07-21
- A coding-agent skill that forces ADHD-friendly, answer-first output — ayghri · 2026-07-21
- A set of agent skills for CAD, robotics, and hardware design — earthtojake · 2026-07-21