Skip to content

v1.3.0 — Terminal-Bench 2.1 result + Claude Opus 5

Choose a tag to compare

View all tags
@ericleepi314 ericleepi314 released this 29 Jul 07:19
· 95 commits to main since this release
v1.3.0
bd3339f
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Highlights

🏆 80.9% on Terminal-Bench 2.1, on Claude Opus 5

Running headless on claude-opus-5 at effort=xhigh, ClawCodex solved 72 of 89 Terminal-Bench 2.1 tasks — 80.9% pass@1 on a single run (k=1). On the public 2.1 leaderboard (k=5 averages) that would slot around third — behind Claude Code / Fable 5 (83.8%) and Codex / GPT-5.5 (83.1%), and ahead of Claude Code on Opus 4.8 (78.9%) and Sonnet 5 (74.6%). Benchmarked on main at #756 (before this tag); run artifacts in eval/harbor/jobs/tb21-clawcodex-3. A directional result, not a ranked submission.

Also in 1.3.0

  • Claude Opus 5 model support, --effort wired end to end (#722, #746)
  • Bounded persistent memory with a background self-improvement review fork (#731)
  • VS Code extension driving the agent-server over stdio (#727)
  • Image-paste input with an [Image #N] un-attach chip (#761, #762)
  • CLAUDE.md → CLAWCODEX.md context-file rebrand, a clean break (#732)
  • Transport-retry hardening so transient network drops resume instead of aborting (#757, #760)

See CHANGELOG.md for the full list.

Install: `curl -fsSL https://clawcodex.app/install.sh | bash` · PyPI: `pip install clawcodex-cli`

Assets 2
Loading