Skip to content

feat(provider): add low reasoning effort for DeepSeek V4 Flash - #224

Merged
Vizards merged 1 commit into
mainfrom
feat/model-specific-reasoning-efforts
Aug 6, 2026
Merged

feat(provider): add low reasoning effort for DeepSeek V4 Flash#224
Vizards merged 1 commit into
mainfrom
feat/model-specific-reasoning-efforts

Conversation

@Vizards

@Vizards Vizards commented Aug 6, 2026

Copy link
Copy Markdown
Owner

Summary

  • model thinking support as per-model capabilities with supported efforts, a default effort, and disable support
  • expose low reasoning effort for DeepSeek V4 Flash while keeping DeepSeek V4 Pro on high and max
  • generate model picker options and validate configured values from the model capability metadata
  • update English and Simplified Chinese labels and documentation

Why

DeepSeek V4 Flash now implements a distinct low reasoning effort, but the extension only exposed none, high, and max. Keeping the effort list in each model definition also avoids hard-coding model-specific behavior in the picker and request pipeline.

none remains a separate UI option that disables thinking rather than an API-level reasoning effort.

User impact

Flash users can select a lightweight reasoning mode for quick edits and simple tasks. Pro users continue to see only the effort levels currently implemented by the official API. Existing and invalid stored values safely fall back to each model's configured default.

Validation

  • npm run format:check
  • npm run compile
  • npm run lint
  • verified in VS Code 1.132 request dumps that Flash main-agent requests send thinking.type: "enabled" with reasoning_effort: "low"
  • verified that internal helper requests continue to use disabled thinking

Closes #220

@Vizards
Vizards marked this pull request as ready for review August 6, 2026 11:44
@Vizards
Vizards requested a balanced review from Copilot August 6, 2026 11:44

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds model-specific reasoning-effort capabilities, exposing low for DeepSeek V4 Flash while preserving Pro’s supported options.

Changes:

  • Centralizes supported and default reasoning efforts in model metadata.
  • Generates picker options and validates configured efforts per model.
  • Updates localized labels and documentation.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated no comments.

Show a summary per file
File Description
src/types.ts Defines reasoning-effort capability types.
src/provider/request.ts Applies validated model-specific effort settings.
src/provider/models.ts Generates picker schemas from capability metadata.
src/i18n.ts Adds localized low labels.
src/consts.ts Configures efforts supported by Flash and Pro.
README.md Documents model-specific effort options.
README.zh-cn.md Updates Simplified Chinese documentation.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@Vizards
Vizards merged commit 3ce1912 into main Aug 6, 2026
2 checks passed
@Vizards
Vizards deleted the feat/model-specific-reasoning-efforts branch August 6, 2026 12:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Flash 用户缺少 low(轻量思考)档位

2 participants