Provider API: GPT-5.6 Luna returns 403 "Authentication failed" despite docs listing it as available on GOAT plan

Summary

The Provider API rejects gpt-5.6-luna with 403 permission_error / "Authentication failed. Please check your credentials.", even though the docs state GPT-5.6 Luna is "Available on every plan, including Go" (https://commandcode.ai/docs/resources/pricing-limits). The same API key works fine with GPT-5.6 Sol, DeepSeek V4, Grok, Kimi, Qwen, GLM etc. google/gemini-3.7-flash exhibits the identical wrong error, while models that genuinely require a higher plan (Terra, Gemini 3.5/3.6, Claude family) correctly return MODEL_NOT_IN_PLAN.

Expected Behavior

gpt-5.6-luna works via Provider API on a GOAT plan (or any plan), as documented on https://commandcode.ai/docs/plans/goat (model table + usage estimates) and https://commandcode.ai/docs/resources/pricing-limits ("Available on every plan, including Go").

Actual Behavior

Every request to gpt-5.6-luna returns:

HTTP 403
{"error":{"message":"Authentication failed. Please check your credentials.","type":"permission_error"}}

Note: permission_error / "Authentication failed" is not listed among the documented Provider API error codes (https://commandcode.ai/docs/provider) — the table there only defines 403 upgrade_required. Reproducible over direct connection and via proxy, from multiple regions, repeatedly (also reproduced in earlier weeks, not just today).

Steps to reproduce the issue

curl https://api.commandcode.ai/provider/v1/chat/completions \
  -H "Authorization: Bearer ***" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-luna","messages":[{"role":"user","content":"hi"}],"max_tokens":16}'

The key IS valid (authenticates fine for other models) and GET /provider/v1/models lists gpt-5.6-luna.

Command Code Version

use API KEY

Operating System

Linux

Terminal/IDE

No response

Shell

No response

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

Docs contradicting observed behavior:

• https://commandcode.ai/docs/plans/goat — Luna in the GOAT model table & usage estimates
• https://commandcode.ai/docs/resources/pricing-limits — "GPT-5.6 Luna … Available on every plan, including Go"; "Gemini 3.7 Flash … Available on GOAT and above"
• https://commandcode.ai/docs/provider — error-code table does not include permission_error

Commencez par reproduire la requête curl vers /provider/v1/chat/completions et comparez-la à celle d’un modèle fonctionnel utilisant la même clé. Lisez ensuite les chemins de listing des modèles de Provider API, de validation du plan et de gestion des erreurs référencés par l’endpoint, ainsi que la documentation liée sur les tarifs et les fournisseurs. Le travail est terminé lorsque les modèles éligibles renvoient la réponse documentée ou l’erreur de plan documentée correcte, plutôt que Authentication failed.