DeepSeekNew

DeepSeek V4.1 Flash 1017 Uncensored

deepseek-v4.1-flash-1017-uncensored

DeepSeek V4.1 Flash 1017 is DeepSeek's fast reasoning model, with thinking effort you can set to low, high or max. This abliterated version answers without refusals, and it is the model our coding agent guides use for Claude Code, Codex and OpenCode. It takes images as well as text. It reads up to 1M tokens of context and writes up to 131K tokens in one reply.

Context
1M
Max output
131K
Input
$0.24 / M
Cached input
$0.038 / M
Output
$0.90 / M

Quickstart

Paste the agent prompt into Claude Code, Codex or any coding agent and it sets everything up. Or use the code directly, with SILK_API_KEY set in your shell.

Set up this project to use the Silk Compute API with the model deepseek-v4.1-flash-1017-uncensored.

Silk Compute is OpenAI compatible:
- Base URL: https://api.silkcompute.com
- Chat: POST /chat/completions (the Anthropic format is also served at /v1/messages)
- Auth: read the API key from the SILK_API_KEY environment variable. Never hard-code or commit it.

Model deepseek-v4.1-flash-1017-uncensored: 1,048,576 tokens of context, up to 131,072 output tokens. Tool calling and streaming are supported.
Reasoning effort (reasoning_effort): low, high, max (default high).

1. Use the OpenAI SDK if this project already has it; otherwise add the official one for its language.
2. Add a small client that sends a chat request to deepseek-v4.1-flash-1017-uncensored and returns the reply text.
3. Add SILK_API_KEY to the example env file if there is one, then show me how to send one test request.

Full API reference: https://silkcompute.com/llms.txt

What is the uncensored DeepSeek API good for?

Start here if you run a coding agent. DeepSeek V4.1 Flash 1017 thinks before it answers, handles tool calls well and reads up to 1M tokens, so Claude Code, Codex or OpenCode can keep most of a repo in view. At the agent mix below, it's the lowest bill of our uncensored models.

Best forCoding agentsTool callsLong contextAll-day runs

What the numbers say. DeepSeek's own table puts 1017 at 42.3 on Terminal-Bench 4.0 and 77.0 on DeepSWE v1.1, up from 31.2 and 74.2 for V4.1 Flash. Those are its numbers for the original model, before abliteration. Does abliteration make it dumber? A little, usually.

What an agent pays. Agents resend their context every turn, so most input is billed at the cached rate of $0.038 per 1M tokens. At the mix our pricing page assumes, 97% cached input, 2% uncached input and 1% output, a million agent tokens cost about $0.05.

How hard it thinks. Set reasoning_effort to low, high or max. High is the default and a good place to stay. Drop to low for simple steps and save max for the hard ones. In Claude Code, switch with /effort. Reasoning effort has the details.

When to step up. For the hardest coding and security work, step up to Kimi K3.1 Pro, our best model for cyber, at about $0.49 per 1M agent tokens. Our coding agent guide sorts every model by job.

Refusal behavior removed, no filter added. The behavior that makes the model refuse is removed from its weights, and we add no content filter on top. Our guide to abliterated models covers what that changes, and the launch post has the rest.

Long agent run tonight? Claude Code needs one settings file. Claude Code setup

Other models

Questions

Is DeepSeek V4.1 Flash 1017 Uncensored really uncensored?

Yes. It is an abliterated version of DeepSeek V4.1 Flash 1017: the behavior that makes the model refuse has been removed from its weights, and Silk Compute adds no content filter on top. It answers the request instead of declining it.

How much does DeepSeek V4.1 Flash 1017 Uncensored cost?

$0.24 per 1M input tokens, $0.038 per 1M cached input tokens, and $0.90 per 1M output tokens, paid from a monthly plan or prepaid credits. Top-ups start at $5.

How do I use DeepSeek V4.1 Flash 1017 Uncensored in Claude Code, Codex or OpenCode?

Point the tool at https://api.silkcompute.com with your Silk Compute API key and set the model to deepseek-v4.1-flash-1017-uncensored. The setup for each tool is on this page and in the docs.

What is the context window of DeepSeek V4.1 Flash 1017 Uncensored?

1,048,576 tokens of context, and up to 131,072 output tokens in one reply.

Does DeepSeek V4.1 Flash 1017 Uncensored take images?

Yes. Send images in the OpenAI, Anthropic or Responses format. Images count as input tokens.

Does Silk Compute store my prompts?

No. Prompts and model outputs pass through our servers and are not stored. We keep only billing records, such as token counts and cost.