spawn_agent model allowlist omits gpt-5.6-luna

Resolved 💬 1 comment Opened Jul 20, 2026 by ohyeh Closed Jul 20, 2026

Codex version

Codex CLI 0.144.6, Codex Desktop on macOS.

What happened

The runtime model catalog exposes gpt-5.6-luna, and the Desktop create_thread / send_message_to_thread schemas also accept it, but the native multi-agent spawn_agent tool rejects the same model:

Unknown model `gpt-5.6-luna` for spawn_agent. Available models: gpt-5.6-sol, gpt-5.6-terra

Reproduction

  1. Confirm the host catalog:
codex debug models | jq -r '.models[].slug'

Key output includes:

gpt-5.6-sol
gpt-5.6-terra
gpt-5.6-luna
  1. In a Codex Desktop task with native collaboration enabled, call spawn_agent with model gpt-5.6-luna and low reasoning.
  2. Observe the allowlist rejection above.

Expected behavior

spawn_agent should accept gpt-5.6-luna when the calling host reports it as available, matching the host model catalog and the other Desktop thread tools. This is useful for low-cost supervision-only sub-agents that mainly execute shell commands, monitor external workers, and summarize progress.

Additional context

gpt-5.6-terra works as a fallback. Using an external codex -m gpt-5.6-luna process is not equivalent because it does not create a native child thread and therefore does not appear in the Codex App Subagents panel.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗