[BUG] ERRO 400 con modelos nuevo como DeepSeek v4,

[awemancba]

Problem (one or two sentences)

"model": "deepseek-v4-pro",

"details": "400\nOpenAI completion error: 400 The `reasoning_content` in the thinking mode must be passed back to the API."

Context (who is affected and when)

cuando uso el modelo pro de deepseek v4, si configuro directo con los modelos viejos de "chat" y "reasoner" funciona solo usando la version flash basica!

Reproduction steps

1. configuro el modelo deepseek v4 pro
2. hago una pregunta
   3? me devuelve el error

Expected result

Que se actualicen con los modelos nuevos de deepseek

Actual result

actualmente no funciona deepseek v4

Variations tried (optional)

No response

App Version

3.53.0

API Provider (optional)

DeepSeek

Model Used (optional)

deepsees-v4-pro

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

I have been looking into this issue and wanted to share some findings and ask for clarification.

Current state of DeepSeek support in Roo Code:

The DeepSeek provider currently supports two model IDs via the official DeepSeek API:

• deepseek-chat (non-thinking mode, maps to DeepSeek-V3.2)
• deepseek-reasoner (thinking mode with reasoning_content support, maps to DeepSeek-V3.2 thinking)

The reasoning_content pass-back (required for thinking/reasoning models) is only enabled when the model ID contains deepseek-reasoner. This is the interleaved thinking mode described in the DeepSeek docs.

About the reported model ID:

The model ID deepseek-v4-pro does not appear to be a standard DeepSeek API model identifier. The official DeepSeek API (at api.deepseek.com) uses deepseek-chat and deepseek-reasoner as model IDs. Could you clarify:

1. Where did you find the model ID deepseek-v4-pro? Is this from a third-party provider or a custom configuration?
2. Are you connecting directly to the DeepSeek API (api.deepseek.com) or through a different base URL / proxy?
3. When you say "if I configure directly with the old chat and reasoner models it works only with the basic flash version" -- could you clarify which exact model IDs work and which do not?

If DeepSeek has introduced new model identifiers (e.g., for a V4 release), we would need to add them to the supported models list and ensure the thinking mode / reasoning_content handling is enabled for those new IDs.

Please tag @roomote in your response with the details, and I will be happy to propose a concrete fix.

[Donghuanyu]

I have been looking into this issue and wanted to share some findings and ask for clarification.

Current state of DeepSeek support in Roo Code:

The DeepSeek provider currently supports two model IDs via the official DeepSeek API:

• deepseek-chat (non-thinking mode, maps to DeepSeek-V3.2)
• deepseek-reasoner (thinking mode with reasoning_content support, maps to DeepSeek-V3.2 thinking)

The reasoning_content pass-back (required for thinking/reasoning models) is only enabled when the model ID contains deepseek-reasoner. This is the interleaved thinking mode described in the DeepSeek docs.

About the reported model ID:

The model ID deepseek-v4-pro does not appear to be a standard DeepSeek API model identifier. The official DeepSeek API (at api.deepseek.com) uses deepseek-chat and deepseek-reasoner as model IDs. Could you clarify:

1. Where did you find the model ID deepseek-v4-pro? Is this from a third-party provider or a custom configuration?
2. Are you connecting directly to the DeepSeek API (api.deepseek.com) or through a different base URL / proxy?
3. When you say "if I configure directly with the old chat and reasoner models it works only with the basic flash version" -- could you clarify which exact model IDs work and which do not?

If DeepSeek has introduced new model identifiers (e.g., for a V4 release), we would need to add them to the supported models list and ensure the thinking mode / reasoning_content handling is enabled for those new IDs.

Please tag @roomote in your response with the details, and I will be happy to propose a concrete fix.

@roomote Regarding your questions:

The model IDs deepseek-v4-flash and deepseek-v4-pro are new models recently released by DeepSeek (or provided via specific OpenAI-compatible endpoints).

The issue is that when using these new IDs in OpenAI compatible mode, the system fails to handle the reasoning_content field, resulting in a 400 Bad Request error: "The reasoning_content in the thinking mode must be passed back to the API."

Please update the provider to treat these new V4 model IDs similarly to deepseek-reasoner so that the thinking/reasoning content is correctly managed in the message history.

I have seen your comment and will explore a potential fix. The issue is that the isThinkingModel check in the DeepSeek provider only matches deepseek-reasoner, so new V4 model IDs like deepseek-v4-flash and deepseek-v4-pro do not get the thinking/reasoning_content handling enabled. I will open a PR to share progress shortly.

Opened PR #12205 with my attempt at addressing this. The fix adds deepseek-v4-flash and deepseek-v4-pro model definitions with thinking mode support, and updates the provider to use the model info preserveReasoning flag instead of hardcoding deepseek-reasoner. Feedback welcome!

[466348376]

所以，现在是等roo官方的接口增加咯？？？请快一点，Roo