Introducing Composer 1.5

Summary

Composer 1.5 is the successor to the first agentic coding model, Composer 1, and is positioned for fast, interactive daily use. It was built by scaling reinforcement learning 20x further on the same pretrained model, with post-training compute exceeding the base model’s pretraining compute. In an internal benchmark of real-world coding problems, performance quickly surpasses Composer 1 and continues rising, with the largest gains on challenging tasks. Composer 1.5 generates thinking tokens to reason about codebases and plan next steps, using minimal thinking on easy problems and continuing until it finds a satisfying answer on hard ones. It also recursively self-summarizes when context runs out, a training behavior intended to preserve original accuracy as context length varies; the release is described as significantly stronger and suitable for interactive use.

Takeaways

1. 1

   Post-training Composer 1.5 used more compute than pretraining the base model, and an internal benchmark reports continued coding gains as reinforcement learning scales.

2. 2

   The model uses minimal thinking for easy problems but continues reasoning on harder problems until it finds a satisfying answer, aiming to preserve interactive speed.

3. 3

   Self-summarization is trained through reinforcement learning and can trigger recursively when context runs out; the source reports maintained accuracy as context length varies.