feat(moonshot): add kimi-k3 and kimi-k2.7-code model support

[lin-snow]

Summary

Add the latest Kimi models to the Moonshot plugin: kimi-k3, kimi-k2.7-code, and kimi-k2.7-code-highspeed.

• kimi-k3: 1M context (1,048,576), vision + tool calling, thinking always enabled. Pricing ¥20/¥100 per 1M tokens (CN platform, cache-miss input / output).
• kimi-k2.7-code / -highspeed: 256K context (262,144), vision + tool calling, thinking mandatory. Pricing ¥6.5/¥27 and ¥13/¥54.

Per official docs, both generations lock temperature / top_p / penalties to fixed values (overriding triggers errors) and do not support non-thinking mode, so these models intentionally expose neither the sampling parameters nor the thinking toggle. llm.py's thinking-model detection is extended to k3 / k2.7 so assistant messages round-trip reasoning_content, which the API requires for multi-turn tool calling.

References: K3 quickstart, K2.7 Code quickstart, K3 pricing, K2.7 Code pricing, Chat API reference

Change Type

• Documentation / non-plugin change
• Non-LLM plugin (tools, extensions, datasource, etc.)
• LLM plugin

Screenshots / Videos

N/A — no K3-enabled API key available; see Testing notes below.

LLM Plugin Checklist

Areas affected by this change (check all that apply)

• Message flow (system messages, user ↔ assistant turn-taking)
• Tool interaction flow (multi-round usage, Agent App and Agent Node)
• Multimodal input (images, PDFs, audio, video, etc.)
• Multimodal output (images, audio, video, etc.)
• Structured output (JSON, XML, etc.)
• Token consumption metrics
• Other LLM functionality (reasoning, grounding, prompt caching, etc.)
• New models / model parameter fixes

Version

• Bumped top-level version in manifest.yaml (not the one under meta) — 0.1.8 → 0.1.9
• dify_plugin is declared in pyproject.toml (>=0.9.0) and locked in uv.lock

Testing

• Local deployment — Dify version:
• SaaS (cloud.dify.ai)

Not runtime-tested: no K3-enabled API key was available. Verification done instead:

• Every field (model IDs, context sizes, max tokens, features, fixed parameters, tool_choice constraints, reasoning_content round-trip requirement, pricing) cross-checked against the official quickstarts, pricing pages, and chat API reference on both platform.kimi.ai (intl) and platform.kimi.com (CN).
• All three model YAMLs load successfully through the SDK's AIModelEntity schema validation.
• tool_choice sent by the OAI-compat SDK is "auto", within kimi-k2.7-code's allowed values (auto / none).

Runtime verification by a reviewer with K3 API access would be appreciated.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[WH-2099]

just wait a moment

[asukaminato0721]

lgtm

[Copilot]

Pull request overview

Copilot reviewed 8 out of 9 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 12 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 12 changed files in this pull request and generated 1 comment.

[WH-2099]

All actionable review threads are resolved. The frozen test suite passes, default runs skip billable live tests, explicit live opt-in passes, and all CI checks are green.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 12 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (1)

models/moonshot/.env.example:1

• Deleting this file removes the standard remote-debug bootstrap for this plugin. The new .difyignore only excludes .env, so it does not make the tracked example redundant; plugins with the same ignore file retain this template (for example, models/aihubmix/.env.example:1-4 and models/tongyi/.env.example:1-4). Restore the template, optionally extending it with the live-test variables.