deepseek-v4-flash missing recommended top_p=0.95 sampling default #40249

Description

opencode does not send a default top_p for deepseek-v4-flash, so requests go out with the API default (1.0). DeepSeek recommends top_p = 0.95 (with temperature = 1.0) for agentic scenarios with this model, and OpenCode is a coding agent:

• https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
• https://docs.api.nvidia.com/nim/re/reference/deepseek-ai-deepseek-v4-flash-infer (defaults to 0.95)

opencode already sets top_p = 0.95 for minimax-m2 and the kimi-k2.5 family, so deepseek-v4-flash should follow the same treatment. The match should be scoped to deepseek-v4-flash only, since other deepseek models (deepseek-chat / reasoner / v3 / v4-pro) have different official guidance.

Plugins

None

OpenCode version

dev branch

Steps to reproduce

1. Run opencode with deepseek-v4-flash
2. Observe the request payload: no top_p is set (falls back to 1.0), while 0.95 is the recommended value for agentic use

Screenshot and/or share link

N/A

Operating System

N/A

Terminal

N/A