feat(xai): update model YAMLs [bot]

[harshiv-26]

Auto-generated by poc-agent for provider xai.

---

Note

Low Risk
Config-only YAML updates that may change model capability gating (e.g., enabling function_calling/structured_output and setting a larger context_window), but no runtime code changes.

Overview
Updates the xAI Grok 4.20 model YAMLs to explicitly declare supported capabilities: both models now list function_calling and structured_output, and grok-4.20-non-reasoning-gv2 additionally enables prompt_caching.

Adds a limits.context_window: 2000000 entry to both model configs, and sets thinking: true on grok-4.20-reasoning-gv2.

^{Reviewed by Cursor Bugbot for commit f5b914d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit f5b914d. Configure here.}

[cursor]

Reasoning gv2 model missing prompt_caching feature

Medium Severity

The grok-4.20-reasoning-gv2 model defines cache_read_input_token_cost in its pricing but omits prompt_caching from its features list. The non-reasoning variant from this same PR (grok-4.20-non-reasoning-gv2) includes prompt_caching, and so do most other grok-4.20 reasoning siblings (grok-4.20-reasoning, grok-4.20-0309-reasoning, grok-4.20-beta-reasoning). This looks like an accidental omission that would cause consumers of this model definition to not know prompt caching is available.

 

^{Reviewed by Cursor Bugbot for commit f5b914d. Configure here.}