Introducing Xploded
Xploded is the MagmaRouter model line: five uncensored models, every one able to call tools, so they can drive a coding agent as well as chat.
Why a short line, not a catalogue
Most uncensored models are roleplay finetunes. They write well and refuse little, but ask one to call a tool and it falls over, which means it cannot read a file, run a command or drive an agent. The models that call tools reliably are the frontier releases, and those are the ones that refuse.
Xploded is the overlap: uncensored or abliterated builds of current frontier models, each one tested with a real tool call before it went on sale. It is five models rather than five hundred because a model only joins the line if it does both jobs.
The models
- GLM 5.3 Xploded (
glm-5.3-xploded): the full GLM 5.3, uncensored, with a million tokens of context. The strongest of the five. - GLM 5.3 Flash Xploded (
glm-5.3-flash-xploded): a million tokens of context, tool calling and vision, at a fraction of the full model's price. The default pick for an agent. - DeepSeek V4.1 Flash Xploded (
deepseek-v4.1-flash-xploded): an uncensored DeepSeek V4.1 Flash, a million tokens of context, tool calling and the longest replies on the line. Text only. Joined on 18 October 2026. - MiMo V2.6 Flash Xploded (
mimo-v2.6-flash-xploded): Xiaomi MiMo with the refusal direction removed, a million tokens of context, tool calling and vision. - Qwen3.8 27B Xploded (
qwen3.8-27b-xploded): small and cheap, half a million tokens of context, the budget pick for agents.
Current prices are on each model page and on pricing, read from the live catalogue.
Uncensored and abliterated are not the same
An uncensored build has been trained to refuse less. An abliterated build has had the refusal direction removed from its weights outright. Neither is a promise that every answer arrives, and both are still capable of being confidently wrong. The difference is explained in full in uncensored vs abliterated.
How to use it
Every Xploded model sits behind one OpenAI-compatible endpoint and one Anthropic-compatible /v1/messages endpoint at https://api.magmarouter.com. Anything that takes a base URL works: the OpenAI SDKs, Cline, OpenCode, SillyTavern, and Claude Code through the Messages endpoint. Name the model by its id and you are done. The integration guides cover each tool.