feat(cli): add kimi-k.3.0 to kimi-coding and related model catalogs

[am423]

Summary

Adds the pre-release Kimi K3.0 model id (kimi-k.3.0) to the Hermes model menus and provider catalogs ahead of its public release.

What changed

• hermes_cli/models.py
  • Added moonshotai/kimi-k.3.0 to OPENROUTER_MODELS and _PROVIDER_MODELS["nous"] so it appears in the Nous portal/OpenRouter curated lists.
  • Added kimi-k.3.0 to _PROVIDER_MODELS for kimi-coding, kimi-coding-cn, moonshot, opencode-zen, opencode-go, alibaba.
  • Added moonshotai/kimi-k.3.0 to _PROVIDER_MODELS for nvidia and novita.
• agent/model_metadata.py
  • Added explicit kimi-k.3.0 context length fallback of 262144.
• plugins/model-providers/kimi-coding/__init__.py
  • Updated default_aux_model from kimi-k2-turbo-preview to kimi-k.3.0 for both the international and China profiles.
• website/static/api/model-catalog.json
  • Regenerated with python3 scripts/build_model_catalog.py so the online manifest matches the in-repo curated lists.
• website/docs/integrations/providers.md and its Chinese translation
  • Updated the quick-start examples to use kimi-k.3.0.
• tests/plugins/model_providers/test_kimi_profile.py
  • Added drift-guard tests asserting kimi-k.3.0 is present in the kimi-coding catalog and accepted by validate_requested_model.

Testing

python3 -m pytest tests/hermes_cli/test_model_validation.py \
                 tests/hermes_cli/test_model_catalog.py \
                 tests/plugins/model_providers/test_kimi_profile.py \
                 tests/agent/test_model_metadata.py -q

Result: 260 passed.

Notes

• OpenRouter has not publicly listed kimi-k.3.0 yet, so the OpenRouter/Nous/Novita/NVIDIA entries are staged for the slug moonshotai/kimi-k.3.0 and can be corrected via the live manifest once the model is released.
• No new provider plugin was needed; this is purely a model catalog update.

Checklist

• Branch cut from latest origin/main.
• Conventional Commits message used.
• In-repo manifest regenerated and matches models.py.
• Targeted tests added and passing.
• Provider docs and i18n docs updated.

[teknium1]

Thanks for the catalog update. Current main does not yet contain this slug, but this version needs availability evidence and a smaller scope before it is safe to salvage.

Problems

• plugins/model-providers/kimi-coding/__init__.py:65,76 makes this pre-release ID the automatic auxiliary model for both endpoints. agent/auxiliary_client.py:469-480,1958-2001 consumes that profile default for side-task requests, while the PR body provides no endpoint-level availability validation.
• website/docs/integrations/providers.md:231,235 promotes the same pre-release ID to both public quick starts.
• tests/plugins/model_providers/test_kimi_profile.py:52,58 are catalog snapshots. AGENTS.md:1321-1340 explicitly rejects exact model-membership change-detector tests.
• agent/model_metadata.py:320 already provides the same 262144 fallback for the entire kimi family, so the new exact entry is redundant.

Suggested changes

• Verify availability per endpoint before changing auxiliary defaults or docs; scope entries to confirmed providers.
• Replace the exact-ID tests with a catalog/manifest invariant, and remove the redundant fallback.

Automated hermes-sweeper review.

[teknium1]

This changes automatic auxiliary requests for kimi-coding, not only the picker: agent/auxiliary_client.py:469-480,1958-2001 selects default_aux_model from the profile. Please retain the existing default until the pre-release slug has been verified against this endpoint (and independently against the China endpoint below).

[teknium1]

Please remove this exact catalog-membership assertion (and the matching validation snapshot below). AGENTS.md:1321-1340 explicitly prohibits change-detector tests for mutable model catalogs; use a catalog/manifest consistency invariant instead.

[teknium1]

Current main already has the generic "kimi": 262144 fallback at agent/model_metadata.py:320, which covers this slug with the same value. This exact entry does not change resolution, so please keep the family mapping as the single source.

[teknium1]

The PR describes this as a pre-release staged ahead of public release. Please do not replace the public quick-start command until this exact slug is verified available on the advertised endpoint; otherwise keep the current known-working example.

[amanning3390]

Superseded by #65725. The replacement is rebased on current main and addresses this review with endpoint-level evidence and strict scope: canonical k3 is discovered dynamically only from https://api.kimi.com/coding/v1/models; legacy/China/custom endpoints filter it, the 1,048,576-token context is endpoint-scoped, and auxiliary defaults, docs, static catalogs, aggregators, and the manifest remain unchanged. 279 focused tests pass.

[teknium1]

Thanks @am423 for flagging Kimi K3 ahead of release — appreciated. Closing this one now that the two halves it covered have landed or been superseded:

• The Nous Portal / OpenRouter curated-list addition landed in feat(models): add moonshotai/kimi-k3 to Nous Portal + OpenRouter curated lists, retire kimi-k2.x #65913 with the final released id moonshotai/kimi-k3 (the pre-release kimi-k.3.0 id in this PR never shipped publicly).
• The kimi-coding endpoint discovery half is superseded by feat(kimi): discover K3 on coding endpoint #65725, which scopes K3 to the endpoint where it's actually served, per review.

Credit to you as the first to submit K3 support.