An uncensored roleplay API with unlimited messages

Uncensored in the weights, not by jailbreak

Most "uncensored" setups are a normal model plus a jailbreak prompt. That works until a scene turns dark, then the model refuses or slides into a lecture mid-story. Unquant's models have refusal behaviour removed in the weights themselves (tensor-level uncensoring), so there is no prompt to maintain and nothing to trip.

Memory that lasts the whole campaign

Character consistency is a memory problem. With a 32K window, a model forgets the promise made forty messages ago. GLM-5.3-Flash serves 262,144 tokens on every plan; Qwen 3.8 Flash Next runs 64K on Roleplay, 128K on Plus and the full 262K from Standard up. See why the context window matters.

Pricing that doesn't punish long stories

Every turn re-sends the whole chat history. On a per-token API a 100K-token story costs as much as a novel every message. Unquant is unlimited: no token caps, no message caps, no meters, one monthly price. Prefix caching also means a long, continuing story starts responding quickly instead of re-reading everything from scratch. The maths is in flat rate vs per-token.

Your client, your character cards

Unquant is a standard OpenAI-compatible API, so it works with the tools roleplayers already use: SillyTavern, Janitor AI (as a proxy), Open WebUI, Agnai, RisuAI and anything else with a custom endpoint setting.

Which plan?

Compare plans · Request beta access