Models
Which model to choose
The API exposes six ids — not six “plans”, but six heads for different jobs. GLM is the default; the others change answer style, text length, and whether reasoning is mandatory. Numbers live on pricing; how to call — in the docs.
Key from @uncensobot, Base URL https://uncensored.monster/v1. Chat Completions only. Empty model = z-ai/glm-5.3-flash-uncensored. The Telegram model picker does not change the API — set the id in your client.
GLM — the default head
z-ai/glm-5.3-flash-uncensored — GLM 5.3 Flash Uncensored. General chat: text, code, and when needed images, search, and reasoning in one id.
Start here if you are unsure, or want one id for the day. Details in the default GLM note. Switch only when the job clearly does not fit: long analytic prose, a different dialog voice, or forced reasoning chains.
Abliterated: short vs heavy text
Two models on the same abliteration line, different scale:
abliteration-ai/abliterated-model— base Abliterated. Short answers, edits, compact drafts.abliteration-ai/abliterated-model-large-v2— Abliterated Large v2. Complex, long text, multi-step write-ups.
Both aim for text with less refusal theater — why that works is in the refusal and abliteration note. Large v2 costs more on input and output; worth it for volume, not for a one-line rewrite.
Qwen: with and without thinking
Same Qwen 3.8 27B base, two ids:
qwen/qwen3.8-27b-uncensored— answer lands in message body, no mandatory “thinking” block.qwen/qwen3.8-27b-uncensored:thinking— same price, but the model builds a reasoning chain before the final answer.
Use :thinking when the path matters: multi-clause logic, option comparison, step plans. Plain Qwen when you want fast chat text or your script only reads content, not a separate reasoning field.
Gemma — dialog and prose
Gemma-4-31B-Gembrain-uncensored-heretic — Gemma 4 31B Gembrain uncensored heretic. Different voice: fiction, roleplay, literary drafts. Not a GLM replacement for code or search — a prose head.
Gemma pricing lists input and output only; no cache line — factor that in vs GLM and Qwen on pricing.
Practical pitfalls
Copy the full id. No [1m] suffix. Client context 1048576, output up to 16384 tokens or your max_tokens. Minimum $0.005 per request — see what a request costs.
Do not hunt for the “most uncensored” id. All six already went through abliteration or equivalent; the split is job and price, not a morality score. Research service — you own the prompt.
Quick map
Daily chat and code → GLM. Short edit → Abliterated base. Long analysis → Abliterated Large v2. Visible reasoning steps → Qwen :thinking. Fast Qwen without thinking → qwen/qwen3.8-27b-uncensored. Creative prose → Gemma.
FAQ
Is an empty model field enough?
Yes — GLM runs. An explicit id is still easier to debug and to A/B heads in one client.
Bot shows Qwen — does the API use Qwen?
Not until the request sets qwen/qwen3.8-27b-uncensored or :thinking. The Telegram picker does not touch the key.
Where is the full price list?
On /en/pricing/. This page is job fit only; per-request math is in the cost note.