feat(microsoft-foundry/gpt-6-astra-2026-09-03): add new models [bot]

[models-bot]

Auto-generated by model-addition-agent for microsoft-foundry/gpt-6-astra-2026-09-03.

---

Note

Low Risk
Metadata-only addition of a model YAML; no runtime or security logic changes.

Overview
Adds a new Microsoft Foundry provider definition for gpt-6-astra-2026-09-03, registering the generally available GPT-6 Astra model for routing, pricing, and capability metadata.

The entry marks the model active and serverless, with chat and responses modes, thinking enabled, and text/image input. It documents tool use (function calling, structured output, prompt caching), a ~1.05M context window, and tiered cumulative token pricing for global and datazone_us (including cache read/write tiers above 272k tokens). Request shaping exposes reasoning_effort (default medium, through max) and verbosity, and explicitly removes temperature and top_p from supported params.

^{Reviewed by Cursor Bugbot for commit 8af0ff3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

/test-models

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 8af0ff3. Configure here.}

[cursor]

Inherited output token cap remains

High Severity

This fuller entry advertises 128000 output tokens but never overrides the inherited max_tokens maxValue of 4096. max_completion_tokens also has no maxValue, and max_tokens is not in removeParams, so the provider default still applies and can be forwarded to a reasoning model that rejects it.

Additional Locations (1)

• providers/microsoft-foundry/gpt-6-astra-2026-09-03.yaml#L49-L51

 

^{Triggered by learned rule: Override max_tokens maxValue to match limits}

^{Reviewed by Cursor Bugbot for commit 8af0ff3. Configure here.}

[cursor]

Unsupported reasoning effort value listed

Medium Severity

reasoning_effort includes none, which GPT-6 Astra does not accept. Official values are low, medium, high, xhigh, and max; sending none returns HTTP 400.

 

^{Reviewed by Cursor Bugbot for commit 8af0ff3. Configure here.}