Skip to content

v5.4.7

Choose a tag to compare

View all tags
@ndycode ndycode released this 22 Mar 06:13
· 122 commits to main since this release
v5.4.7
9dd01d0

Minor Compatibility and Stability Release

This release carries compatibility-facing behavior changes and operational hardening. Read the summary below before upgrading or publishing.

  • Corrects GPT-5 legacy alias routing so gpt-5-mini and gpt-5-nano normalize to the proper GPT-5.4 Mini/Nano families, adds first-class gpt-5.4-nano support, and tightens GPT-5.4 Pro reasoning coercion/validation.
  • Refreshes shipped config metadata for the GPT-5.4 family and Codex models, including the current context-window values, new GPT-5.4 Pro/Nano entries, and updated config template guidance.
  • Preserves caller-supplied max_output_tokens in transformed Responses requests instead of silently clearing explicit output budgets.
  • Treats built-in Responses tools as first-class bridge/runtime entries and expands passthrough coverage for newer Responses API request fields.

Commit Summary

  • 821678b fix: tighten workspace identity follow-ups
  • d9beaab test: close pr86 greptile follow-ups
  • f9db58f Merge pull request #86 from ndycode/fix/pr85-greptile-followups
  • f7d203b fix: correct model alias mapping, add gpt-5.4-nano, validate Pro reasoning
  • be7a969 fix: update config metadata for gpt-5.4 family and Codex context windows
  • 5657d20 fix: preserve max_output_tokens in transformed requests
  • 4de1a27 feat: treat built-in Responses tools as first-class bridge entries
  • 6411373 feat: add prompt cache diagnostics to codex-doctor
  • a9453ac fix: redact prompt cache keys in doctor output
  • 902194e fix: align prompt cache metrics with fetch attempts
  • 4445528 fix: clarify gpt-5 alias comments and pro coercion logs
  • cd6cbc7 docs: clarify max output token passthrough safety
Assets 2
Loading