⚠️ No Changeset found

Latest commit: 809dcc4

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

Greptile Summary

This PR adds two new DeepSeek V4 model definitions (deepseek-v4-pro and deepseek-v4-flash) to the provider and changes the default DeepSeek model from deepseek-chat to deepseek-v4-pro. There are several functional issues that need to be addressed before this is ready to merge:

• isDeepseekV4 is declared in deepseek.ts but never used, leaving the temperature: 0 spread and the addReasoningContent routing unchanged for V4 models — both of which appear to require updating given V4's built-in reasoning/thinking mode.
• Changing the default model ID silently migrates existing users who haven't explicitly chosen a model to a very different model with reasoning enabled and different pricing.

Confidence Score: 2/5

Not safe to merge — multiple P1 logic issues affect runtime behavior for both new and existing users.

Three P1 findings: an unused variable that leaves V4 temperature handling broken, a missing addReasoningContent routing update for reasoning-capable V4 models, and a silent breaking change to the default model for all existing users without an explicit model selection.

Both changed files need attention: deepseek.ts for the unused variable and missing routing logic, and api.ts for the default model change and promotional pricing.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[createMessage called] --> B{model.id includes 'deepseek-reasoner'?}
    B -- yes --> C[openAiMessages with addReasoningContent]
    B -- no --> D[openAiMessages without addReasoningContent]
    D --> E[deepseek-v4-pro / deepseek-v4-flash land here]
    C --> F{model.id === 'deepseek-reasoner'?}
    D --> F
    F -- yes --> G[no temperature param]
    F -- no --> H[temperature: 0 applied]
    H --> I[deepseek-v4-pro / deepseek-v4-flash get temperature:0 ⚠️ isDeepseekV4 never consulted]
    G --> J[API call]
    I --> J

This is a comment left during a code review.
Path: src/core/api/providers/deepseek.ts
Line: 85-99

Comment:
**`isDeepseekV4` declared but never used**

`isDeepseekV4` is defined on line 85 but is never referenced anywhere in the function. The comment on line 98 says "V4 models have thinking on by default," strongly suggesting the intent was to exclude V4 models from the `temperature: 0` spread — just as `deepseek-reasoner` is excluded — but the condition was not updated. As written, V4 models still receive `temperature: 0`, which may conflict with their built-in reasoning/thinking mode.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/shared/api.ts
Line: 2102

Comment:
**Breaking default model change for existing users**

Changing `deepSeekDefaultModelId` from `"deepseek-chat"` to `"deepseek-v4-pro"` silently migrates any user who has not explicitly stored a model selection to the new model. The V4-pro model has `supportsReasoning: true`, a 1M-token context window, and different pricing — a significant behavioral and cost change that users won't be warned about.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/core/api/providers/deepseek.ts
Line: 88-90

Comment:
**V4 reasoning models skip `addReasoningContent` transform**

The new V4 models have `supportsReasoning: true`, but the `openAiMessages` construction only calls `addReasoningContent` for `isDeepseekReasoner` (model ID contains `"deepseek-reasoner"`). If DeepSeek V4's thinking mode produces `reasoning_content` fields in the same way, the input message history will not be properly converted, potentially causing API errors or dropped reasoning context on subsequent turns.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/shared/api.ts
Line: 2104-2114

Comment:
**Hardcoded discounted prices may confuse users**

The `outputPrice`, `cacheWritesPrice`, and `cacheReadsPrice` values for `deepseek-v4-pro` are described as "limited-time 75% off" promotional prices. When the promotion ends, Cline will silently under-bill users. Consider defaulting to the regular prices (with a comment noting the current discount) so cost calculations remain accurate after the promo ends.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "added deepseek-v4 models" | Re-trigger Greptile}

Greptile Summary

This PR adds two new DeepSeek V4 model definitions (deepseek-v4-pro and deepseek-v4-flash) to the provider and changes the default DeepSeek model from deepseek-chat to deepseek-v4-pro. There are several functional issues that need to be addressed before this is ready to merge:

• isDeepseekV4 is declared in deepseek.ts but never used, leaving the temperature: 0 spread and the addReasoningContent routing unchanged for V4 models — both of which appear to require updating given V4's built-in reasoning/thinking mode.
• Changing the default model ID silently migrates existing users who haven't explicitly chosen a model to a very different model with reasoning enabled and different pricing.

Confidence Score: 2/5

Not safe to merge — multiple P1 logic issues affect runtime behavior for both new and existing users.

Three P1 findings: an unused variable that leaves V4 temperature handling broken, a missing addReasoningContent routing update for reasoning-capable V4 models, and a silent breaking change to the default model for all existing users without an explicit model selection.

Both changed files need attention: deepseek.ts for the unused variable and missing routing logic, and api.ts for the default model change and promotional pricing.

Important Files Changed

Filename Overview
src/core/api/providers/deepseek.ts Adds isDeepseekV4 variable but never uses it; the temperature exclusion condition and addReasoningContent routing are not updated to account for V4 models.
src/shared/api.ts Adds deepseek-v4-pro and deepseek-v4-flash model definitions; changes the default model from deepseek-chat to deepseek-v4-pro, which is a silent breaking change for existing users.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[createMessage called] --> B{model.id includes 'deepseek-reasoner'?}
    B -- yes --> C[openAiMessages with addReasoningContent]
    B -- no --> D[openAiMessages without addReasoningContent]
    D --> E[deepseek-v4-pro / deepseek-v4-flash land here]
    C --> F{model.id === 'deepseek-reasoner'?}
    D --> F
    F -- yes --> G[no temperature param]
    F -- no --> H[temperature: 0 applied]
    H --> I[deepseek-v4-pro / deepseek-v4-flash get temperature:0 ⚠️ isDeepseekV4 never consulted]
    G --> J[API call]
    I --> J

Loading

Comments Outside Diff (1)

1. src/core/api/providers/deepseek.ts, line 88-90 (link)
   V4 reasoning models skip addReasoningContent transform
   The new V4 models have supportsReasoning: true, but the openAiMessages construction only calls addReasoningContent for isDeepseekReasoner (model ID contains "deepseek-reasoner"). If DeepSeek V4's thinking mode produces reasoning_content fields in the same way, the input message history will not be properly converted, potentially causing API errors or dropped reasoning context on subsequent turns.
   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: src/core/api/providers/deepseek.ts
   Line: 88-90
   
   Comment:
   **V4 reasoning models skip `addReasoningContent` transform**
   
   The new V4 models have `supportsReasoning: true`, but the `openAiMessages` construction only calls `addReasoningContent` for `isDeepseekReasoner` (model ID contains `"deepseek-reasoner"`). If DeepSeek V4's thinking mode produces `reasoning_content` fields in the same way, the input message history will not be properly converted, potentially causing API errors or dropped reasoning context on subsequent turns.
   
   How can I resolve this? If you propose a fix, please make it concise.
   

Prompt To Fix All With AI

This is a comment left during a code review.
Path: src/core/api/providers/deepseek.ts
Line: 85-99

Comment:
**`isDeepseekV4` declared but never used**

`isDeepseekV4` is defined on line 85 but is never referenced anywhere in the function. The comment on line 98 says "V4 models have thinking on by default," strongly suggesting the intent was to exclude V4 models from the `temperature: 0` spread — just as `deepseek-reasoner` is excluded — but the condition was not updated. As written, V4 models still receive `temperature: 0`, which may conflict with their built-in reasoning/thinking mode.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/shared/api.ts
Line: 2102

Comment:
**Breaking default model change for existing users**

Changing `deepSeekDefaultModelId` from `"deepseek-chat"` to `"deepseek-v4-pro"` silently migrates any user who has not explicitly stored a model selection to the new model. The V4-pro model has `supportsReasoning: true`, a 1M-token context window, and different pricing — a significant behavioral and cost change that users won't be warned about.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/core/api/providers/deepseek.ts
Line: 88-90

Comment:
**V4 reasoning models skip `addReasoningContent` transform**

The new V4 models have `supportsReasoning: true`, but the `openAiMessages` construction only calls `addReasoningContent` for `isDeepseekReasoner` (model ID contains `"deepseek-reasoner"`). If DeepSeek V4's thinking mode produces `reasoning_content` fields in the same way, the input message history will not be properly converted, potentially causing API errors or dropped reasoning context on subsequent turns.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/shared/api.ts
Line: 2104-2114

Comment:
**Hardcoded discounted prices may confuse users**

The `outputPrice`, `cacheWritesPrice`, and `cacheReadsPrice` values for `deepseek-v4-pro` are described as "limited-time 75% off" promotional prices. When the promotion ends, Cline will silently under-bill users. Consider defaulting to the regular prices (with a comment noting the current discount) so cost calculations remain accurate after the promo ends.

How can I resolve this? If you propose a fix, please make it concise.

Reviews (1): Last reviewed commit: "added deepseek-v4 models" | Re-trigger Greptile

fixed