DeepSeek V4-Pro sometimes returns no visible response after API reasoning completes

[tanfeifan]

Description

Bug Description

When using DeepSeek V4-Pro in OpenCode, the model occasionally enters the reasoning/thinking phase and the API request appears to complete successfully, but no final response is displayed in the chat output.

Expected Behavior

After the reasoning process finishes, the model response should be returned and rendered normally in the conversation.

Actual Behavior

The API call seems to execute normally and the model spends tokens on reasoning, but the final assistant message is missing or empty.

Possible Symptoms

• The request shows as completed
• Tokens are consumed
• Thinking/reasoning logs may appear
• No assistant reply is rendered
• Sometimes the issue occurs randomly and cannot be reproduced consistently

Environment

• OpenCode
• Model: DeepSeek V4-Pro

Additional Notes

This issue appears intermittently and may be related to streaming, response parsing, or handling of reasoning outputs.

Plugins

zero

OpenCode version

All software versions that have been used have this problem, including the latest one.

Steps to reproduce

No response

Screenshot and/or share link

There is content between the two sentences I sent.

Operating System

win11

Terminal

No response

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Provider stream truncation (finishReason="other" with zero output) silently accepted, persisting half-finished assistant messages #26170: Provider stream truncation silently accepted — API call completes but response is half-finished or empty, very similar symptom
• Silent completion failure for non-default FOSS models on OpenRouter (Kimi K2.6, DeepSeek-R1, llama-3.3-70b) #26244: Silent completion failure where model returns no visible response through opencode's adapter layer (affects DeepSeek-R1 among others)
• deepseek V4 flash 或者pro 出现 链接中断 ，中断后 再也不能正常对话工作了Error from provider (DeepSeek): The reasoning_content in the thinking mode must be passed back to the API. #24920: DeepSeek V4 Pro / Flash specifically — reasoning interruptions and missing responses after tool calls

[tanfeifan]

If "Show Reasoning Content" is enabled, it can be seen that the API actually responded correctly.

[ListenLove]

This issue affected my work severely, I've checked out to deepseek official API temporarily.