DeepSeek V4.1 Flash RR API: abliterated DeepSeek V4.1 Flash 1017

By the RedRouter Team · Tested October 27, 2026
September 17, 2026

Aircraft test model mounted inside the NASA Ames 12-foot wind tunnel, overlaid with red pixel blocks and the title DeepSeek V4.1 Flash RR — Cover based on a photo by NASA Ames Research Center, AC95-0203-37, Public domain (NASA)

DeepSeek V4.1 Flash RR is an abliterated version of DeepSeek's V4.1 Flash, built from the 1017 checkpoint. It has a million-token context and handles long-context chat, coding and tool use. The model ID is deepseek-v4.1-flash-rr.


Model ID	`deepseek-v4.1-flash-rr`
Context window	1,048,576 tokens
Maximum output	8,192 tokens
Default output	1,024 tokens
Tool calling	Yes
JSON response format	Yes
Quantization (host-reported)	FP8
Price per 1M tokens	$1.15 input, $0.115 cached input, $2.30 output

Zero Data Retention, if you turn it on for your account, adds 20% to every rate.

Abliterated, not fine-tuned

Four steps from the DeepSeek V4.1 Flash base model to the RedRouter model ID deepseek-v4.1-flash-rr: modified weights with unknown revision, a host serving them, and one pinned RedRouter deployment — From the base model to deepseek-v4.1-flash-rr, and what is known at each step. Image: RedRouter, Original

Abliteration removes the direction in the model's activations that drives refusals, without retraining it on new data. The host identifies the base as the 1017 release of DeepSeek V4.1 Flash, published October 17, 2026. As of October 27 it is the newest V4.1 Flash checkpoint in DeepSeek's model listing.

What is not verified: the modified weights' exact revision, license, tokenizer and chat template. The catalog lists each as unknown. The host reports FP8 weights.

DeepSeek's published results for the base checkpoints, not for the RedRouter modified version. RedRouter has not reproduced them.

Benchmark	V4.1 Flash 1017	Earlier V4.1 Flash
Terminal-Bench 4.0	42.3	31.2
DeepSWE v1.1	77.0	74.2
ProgramBench	29.1	20.3
NL2Repo-Bench	68.2	65.4

Source: DeepSeek V4.1 Flash 1017 model card, read October 27, 2026. HunterBench lists DeepSeek V4 Flash, a different model, so its score is not repeated here.

Try it with tools

curl https://redrouter.org/v1/chat/completions \
  -H "Authorization: Bearer $REDROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash-rr",
    "messages": [{"role": "user", "content": "Look up CVE-2024-3094."}],
    "tools": [{
      "type": "function",
      "function": {
        "name": "lookup_cve",
        "description": "Look up a CVE by ID.",
        "parameters": {
          "type": "object",
          "properties": {"id": {"type": "string"}},
          "required": ["id"]
        }
      }
    }],
    "tool_choice": "auto"
  }'

The response carries a tool_calls entry. Your code runs the function and sends the result back in a tool message. RedRouter never executes tools.

Run on October 27, 2026. deepseek-v4.1-flash-rr returned a tool_calls entry:

[
  {
    "function": {
      "arguments": "{\"id\": \"CVE-2024-3094\"}",
      "name": "lookup_cve"
    },
    "id": "call_7e7f12d1a30b4520b6e58fa247d58e96",
    "type": "function"
  }
]

When to pick it

It costs more than GLM 5.3 Flash RR: $1.15 against $0.20 input, $2.30 against $0.80 output per million tokens. Pick DeepSeek V4.1 Flash RR when you want its abliterated 1017 checkpoint specifically, or when its answers fit your task better in your own testing. It carries the full million-token context, tool calling and JSON mode. When cost matters most, use GLM 5.3 Flash RR.

FAQ

Can I use it in Claude Code or OpenCode? Yes. See the agent setup guide.

Is my data kept? By default, an encrypted copy of prompts and completions is kept for up to 30 days for abuse prevention, support, debugging and billing disputes. Zero Data Retention stops that. See the privacy policy.

What may I use it for? Anything legal. Testing systems you do not own or are not authorized to test is against the terms.