DeepSeek Flash Lite Xploded

The abliterated DeepSeek Flash Lite API

MagmaRouter serves an abliterated build of DeepSeek Flash Lite under the model id deepseek-v4.1-flash-lite-1017-xploded. It has 512k tokens of context, costs $0.36 per million input tokens and $1.44 per million output tokens, with cached input at $0.03, and supports tool calling, so it can drive an agent. One key works on the OpenAI-compatible endpoint (https://api.magmarouter.com/v1) and on the Anthropic-compatible one Claude Code uses, so the same DeepSeek Flash Lite model plugs into Claude Code, Cline, opencode, SillyTavern, the OpenAI SDK and anything else that takes a custom base URL.

DeepSeek Flash Lite abliterated against the rest of the line

Prices are per million tokens. Intelligence is the Artificial Analysis Intelligence Index of the base model. DeepSeek V4.1 Flash 1017 Xploded has no score of its own, so it shows DeepSeek V4.1 Flash's, the model it was tuned from. DeepSeek Flash Lite Xploded has no score of its own, so it shows DeepSeek V4.1 Flash's, the model it was tuned from.

Uncensored DeepSeek V4.1 Flash sets it against the rest of the line, and uncensored DeepSeek goes through the DeepSeek releases one by one.

What the abliterated tag actually means

deepseek-v4.1-flash-lite-1017-xploded is tagged abliterated, meaning the refusal direction was identified and removed from the model's activations, so it very rarely refuses. That is not the same as being correct or safe: it can still be wrong, produce weak output, or generate content that is illegal to use. The tag describes what was done to the weights, not a promise about answers.

Call it with curl

curl https://api.magmarouter.com/v1/chat/completions \
  -H "Authorization: Bearer sk-mr-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash-lite-1017-xploded",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Call it with the OpenAI SDK

# pip install openai
from openai import OpenAI

client = OpenAI(
    base_url="https://api.magmarouter.com/v1",
    api_key="sk-mr-...",
)

resp = client.chat.completions.create(
    model="deepseek-v4.1-flash-lite-1017-xploded",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

In Claude Code

This model supports tool calling, so it can drive Claude Code. Set the context length to its real value of 512,000:

# Put these in your shell profile, then run: claude
export ANTHROPIC_BASE_URL=https://api.magmarouter.com
export ANTHROPIC_AUTH_TOKEN=sk-mr-...
export ANTHROPIC_MODEL=deepseek-v4.1-flash-lite-1017-xploded
export ANTHROPIC_DEFAULT_HAIKU_MODEL=glm-5.3-xploded
export CLAUDE_CODE_MAX_CONTEXT_TOKENS=512000

FAQ

The rest of the line