[privatemode] sync supported models

[jmxnzo]

Generated by the Privatemode sync-models-dev workflow.

This PR synchronizes the Privatemode model definitions in models.dev with the current production deployment.

The workflow is triggered when:

• a model is deprecated
• a new model is added
• an existing model is updated

Please visit the Privatemode model overview for more information about the supported models.
Check the release notes for the latest model updates and changes.

[github-actions]

Action items

• [high] [violation] providers/privatemode-ai/models/gemma-4-31b.toml:3 - Check: base_model must resolve to an existing models/ entry. Why: The value google/gemma-4-31B-it does not match any metadata file; the actual file is models/google/gemma-4-31b-it.toml (lowercase b). generate() keys the metadata map by the case-sensitive filesystem path, so bun validate throws Unable to resolve base_model and the whole catalog fails to build. Action: Change to base_model = "google/gemma-4-31b-it".
• [high] [violation] providers/privatemode-ai/models/kimi-k2.6.toml:3 - Check: base_model must resolve to an existing models/ entry. Why: The value moonshotai/Kimi-K2.6 does not match any metadata file; the actual file is models/moonshotai/kimi-k2.6.toml (lowercase k). Same case-sensitivity failure as above. Action: Change to base_model = "moonshotai/kimi-k2.6".
• [high] [violation] providers/privatemode-ai/models/qwen3-embedding-4b.toml:3 - Check: base_model must point to an existing models/<provider>/<model>.toml; otherwise a full inline definition is required. Why: Qwen/Qwen3-Embedding-4B resolves to nothing — there is no models/Qwen/ directory and no Qwen3-Embedding metadata anywhere in models/. The file also no longer carries release_date/last_updated/description, so even if the pointer were ignored it would fail the required-fields schema. Action: Restore a full inline provider definition (with description, release_date, last_updated, etc.) or add the corresponding models/ metadata entry in this PR.
• [high] [violation] providers/privatemode-ai/models/voxtral-mini-3b.toml:3 - Check: base_model must point to an existing models/ entry; otherwise a full inline definition is required. Why: mistralai/Voxtral-Mini-3B-2507 resolves to nothing — the directory is models/mistral/ (not mistralai) and no Voxtral metadata file exists. The file is also missing release_date/last_updated/description. Action: Restore a full inline provider definition (with description, release_date, last_updated, etc.) or add the corresponding models/ metadata entry in this PR.
• [medium] [possible mistake] providers/privatemode-ai/models/gemma-4-31b.toml:6 and providers/privatemode-ai/models/kimi-k2.6.toml:6 - Check: reasoning_options must reflect the inference provider's actual API surface; niche providers documenting a toggle must express the exact request syntax as an adjacent TOML comment. Why: Both new files declare reasoning_options = [{ type = "toggle" }] with no comment, while providers/privatemode-ai/provider.toml states Privatemode "defines no provider-wide reasoning enum or bounds" and only passes parameters through to the active model server. There is no evidence in the PR that Privatemode exposes a verified reasoning toggle for these models. Action: Either add a comment documenting the exact API request syntax for the toggle (e.g. # API: {"chat_template_kwargs": {"enable_thinking": false}}) and confirm the provider supports it, or use reasoning_options = [] if no caller control is verified.
• [low] [possible mistake] providers/privatemode-ai/models/whisper-large-v3.toml:17 - Check: Material factual changes should be supported by a citation. Why: limit.output changes from 4096 to 0 with no source, and the PR body provides no citations for any data change. An output limit of 0 for a transcription model that emits text is unusual and cannot be reviewed against evidence. Action: Provide a Privatemode API/docs reference for the output limit (or revert to the prior value if 0 is not intentional).

[github-actions]

Action items

• [critical] [violation] providers/privatemode-ai/models/gpt-oss-120b.toml:1 - Check: base_model must be retained when a matching models/ entry exists. Why: The diff removes base_model = "openai/gpt-oss-120b" and inlines provider-agnostic fields, but drops the inherited description, release_date, and last_updated without re-adding them. The resulting file has no description/release_date/last_updated and no base_model to inherit from, so it fails the required-field schema and breaks catalog generation. Action: Restore base_model = "openai/gpt-oss-120b" and keep only provider-specific fields (the name/release_date/last_updated/knowledge overrides, reasoning_options, [cost], and the limit.output = 131_072 override).

• [critical] [violation] providers/privatemode-ai/models/kimi-k2.6.toml:3 - Check: base_model must point to an existing models/ entry with the correct case. Why: The file sets base_model = "moonshotai/Kimi-K2.6" (uppercase K), but the metadata file is models/moonshotai/kimi-k2.6.toml (lowercase). On the case-sensitive checkout, resolveBaseModel throws "Unable to resolve base_model"; every other provider uses moonshotai/kimi-k2.6. Action: Change to base_model = "moonshotai/kimi-k2.6".

• [critical] [violation] providers/privatemode-ai/models/qwen3-embedding-4b.toml:3 - Check: base_model must point to an existing models/ entry. Why: The diff adds base_model = "Qwen/Qwen3-Embedding-4B", but no models/Qwen/ namespace exists and no Qwen3-Embedding metadata is present anywhere under models/. resolveBaseModel will throw, and the previously inline description/release_date/last_updated were deleted so nothing supplies them. Action: Either create the missing models/Qwen/Qwen3-Embedding-4B.toml metadata or revert to a full inline definition that includes description, release_date, and last_updated.

• [critical] [violation] providers/privatemode-ai/models/voxtral-mini-3b.toml:1 and providers/privatemode-ai/models/whisper-large-v3.toml:2 - Check: Provider models without base_model must include all required fields. Why: Both files lack base_model and omit the required description, release_date, and last_updated — whisper had them before and the PR deleted them; voxtral never had them. No inheritance supplies them, so both fail validation. Action: Restore/add the required fields inline, or add a valid base_model (e.g., openai/whisper-large-v3 for whisper; for voxtral, create models/mistral/voxtral-mini-3b.toml first or inline the fields).

• [high] [violation] providers/privatemode-ai/models/gemma-4-31b.toml:1 and providers/privatemode-ai/models/kimi-k2.6.toml:1 - Check: Provider-agnostic facts must not be duplicated inline when base_model is used. Why: Both files set base_model yet re-declare attachment, reasoning, structured_output, temperature, tool_call, open_weights, family, modalities, and limit.context identically to the metadata. The checklist marks this as a blocker. Action: Keep only provider-specific fields and genuine overrides (e.g., limit.output for gemma; modalities.input/family for kimi; reasoning_options; [cost]).

• [medium] [possible mistake] providers/privatemode-ai/models/gemma-4-31b.toml:4 and providers/privatemode-ai/models/kimi-k2.6.toml:4 - Check: reasoning_options toggle claims require provider evidence and a TOML comment documenting the request syntax. Why: Both files declare reasoning_options = [{ type = "toggle" }] with no comment showing the API request field/values, while provider.toml states Privatemode "defines no provider-wide reasoning enum or bounds." The audit skill requires a documented or reproduced toggle mechanism before claiming toggle. Action: Either add a TOML comment with the exact API request syntax and cite evidence in the PR body, or change to reasoning_options = [].

• [low] [possible mistake] providers/privatemode-ai/models/qwen3-embedding-4b.toml:17 and providers/privatemode-ai/models/whisper-large-v3.toml:14 - Check: Factual limit values need support. Why: limit.context and limit.output are set to 0 for an embedding model (which accepts input tokens) and a transcription model (which emits text), with no citation; the prior non-zero values were overwritten. Action: Provide a source for the zero limits or restore the prior values.

• [low] [violation] pull-request.json:4 - Check: Data-changing PRs should cite direct provider sources. Why: The PR body gives no pricing, model documentation, or API references for the new models, deletions, reasoning-option claims, or limit changes. Action: Add first-party citations stating what each link supports (model docs, API reference, pricing).