Agent Zero v2.7 is out.
- Proxy support for the internal browser, per project, with auth and bypass options.
- Cerebras joins as a provider, if you want some of the fastest inference around driving your agents
(video below).
More on GitHub changelog 👇
Claude Code vs Agent Zero, same model:
Claude Opus 5 High, released yesterday.
Build a working Web OS from scratch.
Terminal, apps, two games.
Claude Code: 9:58, 80.3k tokens
Agent Zero: 11:34, 59.7k tokens
That's 26% fewer tokens for the same result.
Full test at the end
Agent Zero crushed Codex by 2.6× on token efficiency!
We gave both agents the same task on the same local model (Qwen3.6 27B): build a retro Q*bert arcade game. Each agent wrote and rewrote it across iterations, adding enemies and a HUD.
Outputs:
• Agent Zero: 239K tokens, 53 min
• Codex: 627K tokens, 51 min
Both agents built a working Q*bert, but @gitlawb 's Agent Zero is the cleaner build, with crisper textures and hops that land where they should. Codex refills the context window with its whole history every turn, so the model loses the thread and hallucinates code it never wrote. Agent Zero sends only what's new, so it keeps more of the window open and makes fewer mistakes.
Run agents on local AI models in Atomic Chat!
Thanks for the comparison.
But the Agent Zero you use in this video is not made by gitlawb, it’s the original we have developed 😉
We work hard to make your Qbert efficient 🥹
Agent Zero v2.6 is out, plus a fresh Launcher update.
The big one: no more JSON tool misformats.
The Launcher now walks you through staged Accessibility permissions for Computer Use, lets you choose per Instance, and recovers cleanly after restarts.
You don't need our form to create a Project in Agent Zero.
Just ask in plain English, and watch the agent build its own terminal tool to create the project. In real time. In under a minute.
Free your hands for other work. Agent Zero takes care of the rest. (gpt-5.6-terra)