Bullet: a free coding agent claiming 30–60% faster than Claude Code
Free, and it rides your Claude Code or Codex login — auto-routing models per prompt, parallelizing reads, and posting a top-3 SWE-bench Verified score.
Copy markdown
30–60% faster, with a top-3 SWE-bench score to show
Bullet says it cuts 30–60% off wall-clock time versus Claude Code and Codex on the same work, and lists 95.8% on SWE-bench Verified at a 119-second average task time — a top-3 result. The benchmark is concrete; treat the speed number as a vendor claim until independent runs confirm it.
It runs on the subscription you already pay for
Rather than reselling tokens, Bullet connects over OAuth to your existing Claude Code or Codex plan — or your own API keys, or on-device models — and the agent itself is free. If you already hold a Max or Codex plan, trying it costs nothing new.
The speed is orchestration, not a bigger model
Bullet auto-picks the model and reasoning level per prompt, runs searches, file reads and shell commands in parallel, and does targeted code search instead of embedding your whole repo. It's a harness play — routing and parallelism, not a new frontier model.
CLI and GUI, free to try this weekend
It ships as a terminal CLI and a desktop GUI at codewithbullet.com, from a two-person team of Yale CS grads formerly at AppLovin and Citadel. It's early, so expect rough edges — but the zero-cost, bring-your-own-login setup makes it a low-risk swap-in on a real task.