DeepSeek V4.1 Flash 1017 Uncensored is live

DeepSeek V4.1 Flash 1017 Uncensored is live on Silk Compute. It is DeepSeek's reasoning model with its refusal behavior removed, and it is the model our coding agent guides use.
What is it good at?
Don't let the word Flash fool you. At 763B parameters, this is a very large model, and long coding sessions are where we'd use it. It thinks before it answers, it handles tool calls well, and it reads up to 1M tokens, so a coding agent can keep most of a repo in view.
DeepSeek calls 1017 a minor update, but its own table shows a real jump over V4.1 Flash: 42.3 against 31.2 on Terminal-Bench 4.0, and 77.0 against 74.2 on DeepSWE v1.1. Those are its numbers for the original model, before abliteration.
Which reasoning effort should you use?
You choose how hard it thinks: set reasoning_effort to low, high or max. High is the default, and it's a good place to stay. Drop to low for simple steps, and save max for the hard ones.
What does it cost in a coding agent?
About $0.05 per 1M tokens today, at the mix on our pricing page: 97% cached input, 2% uncached input, 1% output. That mix is an assumption. It's the lowest bill of our five models at that mix, which is why it's where we'd start. Our coding agent guide has the full comparison.
Specs and pricing
| DeepSeek V4.1 Flash 1017 | |
|---|---|
| Model id | deepseek-v4.1-flash-1017-uncensored |
| Context | 1,048,576 tokens |
| Max output | 131,072 tokens |
| Input | $0.24 per 1M tokens |
| Cached input | $0.038 per 1M tokens |
| Output | $0.90 per 1M tokens |
| Reasoning effort | low, high, max (default high) |
Coming from DeepSeek V4 Flash?
deepseek-v4.1-flash-1017-uncensored replaces deepseek-v4-flash-uncensored, which now returns model_not_found. Change the model id and nothing else.
How do I get started?
Create an API key and send a request to deepseek-v4.1-flash-1017-uncensored at https://api.silkcompute.com. The model page has a prompt you can paste into your coding agent to set it up for you, plus code for Python, JavaScript, cURL, Claude Code, Codex and OpenCode.


