Across our benchmarks, the model sets a new standard.
It scores 52.6% on Terminal-Bench-Science 0.1, more than double Fable 5. On Terminal-Bench 4.0, it scores 55.8% against 42.0% for Fable 5.
Spent the morning inside Berd, @blocks newly open-sourced agent workspace, and the architecture is the story.
It ships no model of its own. It speaks ACP - Zed's open standard, LSP-but-for-agents - so Claude Code, Codex, Cursor and Goose are all just interchangeable backends.
