Launching today
AgentCodec
Combine cheaper models to rival frontier LLMs and VLMs
1 follower
Combine cheaper models to rival frontier LLMs and VLMs
1 follower
AgentCodec combines any mix of LLMs and VLMs, hosted or open-weight, using techniques inspired by communication theory and more: retry with feedback, add redundancy, merge several outputs. Even two open-weight models combined clearly improve benchmark results. Evaluate 28 techniques on your own data, then a router picks the model and technique for each request, with one knob for cost vs. quality. One OpenAI-compatible endpoint. Built on our NeurIPS 2026 paper, and extended since.
