🤖 bench: target GPT-5.6 Sol and run Terminal-Bench lanes at high thinking

[ibetitsmike]

Summary

Updates the Terminal-Bench model targets: the nightly GPT lane moves from GPT-5.5 to GPT-5.6 Sol, and all benchmark lanes now run at --thinking high (Claude Opus 5 previously ran xhigh).

Background

GPT-5.6 Sol replaced GPT-5.5 as the flagship OpenAI tier, and the requested bench configuration is GPT-5.6 Sol at high thinking and Claude Opus 5 at high thinking. The old per-model xhigh/high conditional in the nightly workflow is no longer needed, so it is flattened to a single --thinking high.

Implementation

• nightly-terminal-bench.yml: default all matrix swaps openai/gpt-5.5 for openai/gpt-5.6-sol; mux_run_args is now a plain --thinking high for every lane. Gemini lanes and the Sonnet smoke tests are unchanged.
• terminal-bench.yml: model example strings refreshed.
• prepare_leaderboard_submission.py: adds leaderboard metadata for anthropic/claude-opus-5 (previously missing, so submission prep could not map the current Opus target) and openai/gpt-5.6-sol; older entries stay as historical mappings.
• tbench skill (and its benchmarks/terminal_bench/README.md symlink): example commands updated to the new targets.

Validation

• make static-check green locally.
• YAML parse of both workflows and py_compile of the leaderboard script pass.
• Grep confirms the only remaining gpt-5.5 references are intentional historical metadata entries.

---

Generated with mux • Model: anthropic:claude-fable-5 • Thinking: xhigh • Cost: $5.53

[ibetitsmike]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Bravo.

Reviewed commit: 6cf31f7c99

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".