Kimi K3.1 Pro, uncensored, is on MagmaRouter

Kimi K3.1 Pro, uncensored, is on MagmaRouter

Kimi K3.1 Pro Xploded is an abliterated Kimi K3.1 Pro: 1,048,576 tokens of context, tool calling, reasoning and images. Model id kimi-k3.1-pro-xploded, $7.50 in and $37.50 out per million tokens.

What it is

Kimi K3.1 Pro Xploded (kimi-k3.1-pro-xploded) is Moonshot AI's Kimi K3.1 Pro with the refusal direction taken out of its weights. Kimi K3.1 Pro launched fourth on the Artificial Analysis Intelligence Index, at 55 on v4.3.2, behind only Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full. It went out without weights; they are public now, and that is what made an abliterated build possible.

It reads 1,048,576 tokens of context, writes up to 943,718 in one reply, calls tools, reasons before it answers and reads images. It is the strongest model on the line.

What it costs

$7.50 per million input tokens, $0.80 per million cached input tokens and $37.50 per million output tokens. A 25-turn coding agent session that re-reads 750,000 tokens of context, 85% of it cached, and writes 25,000 costs about $2.29. The full breakdown is on Kimi K3.1 Pro pricing.

Calling it

Any OpenAI client works with the base URL https://api.magmarouter.com/v1 and the model id kimi-k3.1-pro-xploded:

curl https://api.magmarouter.com/v1/chat/completions \
  -H "Authorization: Bearer $MAGMAROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi-k3.1-pro-xploded", "messages": [{"role": "user", "content": "Hello"}]}'

Anthropic clients, Claude Code among them, use https://api.magmarouter.com with the same key. Setups: Claude Code, Codex CLI, the OpenAI SDK.

When to use it

Use it for the hardest steps: planning a large change, debugging what a cheaper model could not, long reports and whole generated files. For routine agent steps GLM 5.3 Xploded costs $2.00 in and $5.00 out per million, on the same key. The two side by side: Kimi K3.1 Pro vs GLM 5.3, and against the line's value pick, Kimi K3.1 Pro vs DeepSeek V4.1 Flash 1017.

More from the blog