Imported from https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash.

Requires ollama v0.31.1+.

This model needs a parser/renderer to implement the encoding that the model uses. The deepseek3.1 parser/renderer doesn’t seem to work very well, so this model has a Jinja template. For that reason, it’s tool call abilities may not be as good as they could be.

Chat/thinking

$ ollama run frob/deepseek-v4-flash
>>> hello, who are you?
Thinking...
The user is asking for my identity. I should respond in a friendly and informative 
manner.
...done thinking.

Hello! I'm your helpful AI assistant. You can think of me as a smart digital 
companion designed to answer questions, provide information, assist with tasks, 
and have conversations just like this one. How can I help you today?

>>> /clear
Cleared session context
>>> /set nothink
Set 'nothink' mode.
>>> hello, who are you?
Hello! I'm DeepSeek, an AI assistant created by the company DeepSeek (深度求索). 
I'm here to help answer your questions, assist with tasks, provide information, 
and have conversations with you.

I can help with a wide range of topics—from answering general knowledge questions 
to helping with writing, coding, analysis, and more. If there's anything specific 
you'd like to know or discuss, feel free to ask!

What brings you here today? 😊

Tools

$ ollama launch opencode \
    --model frob/deepseek-v4-flash \
    -- \
    run 'create a file called README with the contents "hello world"'

> build · frob/deepseek-v4-flash

← Write README.md
Wrote file successfully.

I'll create the README file for you.

Done. Created `issues/dsv4flash/README.md` with contents "hello world".

$ cat README.md 
hello world