feat(deepinfra): update model YAMLs [bot]

[harshiv-26]

Auto-generated by poc-agent for provider deepinfra.

---

Note

Medium Risk
Primarily configuration-only YAML updates, but changes to max_*_tokens, context_window, and declared features (e.g., json_output) can alter runtime behavior and token budgeting for callers.

Overview
Updates DeepInfra provider model YAMLs to reflect current capabilities and metadata: adds status: active broadly, introduces/expands modalities sections, and annotates more models with features like json_output, structured_output, and function_calling.

Adjusts several token limits/pricing knobs (e.g., adds missing max_output_tokens, reduces max_tokens for some Qwen models, increases context_window for Kimi models, adds tiered cache-read pricing for Qwen3-Max) and marks meta-llama/Llama-Guard-3-8B as deprecated via isDeprecated + deprecationDate.

^{Written by Cursor Bugbot for commit e23d6c7. This will update automatically on new commits. Configure here.}

[cursor]

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

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

[cursor]

Limits section accidentally removed from Janus-Pro-1B

High Severity

The entire limits section (including max_input_tokens: 2600) was removed from the Janus-Pro-1B config. The sibling model Janus-Pro-7B.yaml retains its limits with max_input_tokens: 2600. Removing input token constraints means clients won't know the model's actual input limit, potentially leading to failed API calls when exceeding the undocumented cap.

 

[cursor]

Redundant supportedModes field added to deepinfra model

Low Severity

A supportedModes field with - chat was added to this deepinfra model config, but this field is not used in any other deepinfra model. It only appears in azure-open-ai provider configs. The existing mode: chat field on line 24 already conveys the same information, making supportedModes redundant and inconsistent with the rest of the deepinfra provider.