Infermatic.ai is a flat-rate host for creative writing and roleplay fine-tunes, served “without guardrails”. It’s popular with SillyTavern users, and every paid plan includes an OpenAI-compatible API. It doesn’t claim to serve abliterated or uncensored weights: the models are community roleplay fine-tunes. We list 9 models, all fine-tuned.
This review is based on Infermatic’s published plans, model list, and policies, checked on October 11, 2026. Benchmark results from our own runs will be added when they’re complete. Live details are on the Infermatic provider page.
The short version
| Best for | Heavy roleplay or story-writing from SillyTavern or another front-end, for a fixed monthly price |
| Uncensored models | 9 (all fine-tuned) |
| Per-token price | None: flat rate |
| Plans | Free (web chat only), Essential $9/mo, Standard $16/mo, Plus $20/mo |
| Privacy score | 40 / 100, joint 14th of 37 |
| Watch out for | Rate limits of 12 to 18 requests a minute, and nothing above 72B is listed yet |
Models
Infermatic’s catalog is a who’s-who of community roleplay fine-tunes:
- 70B to 72B: Magnum v4 SE (Llama 3.3 70B), Euryale v2.3 (Llama 3.3 70B), Hanami x1 (Llama 3.1 70B), Anubis 70B v1.1 (Llama 3.3 70B), Strawberrylemonade v1.1 (Llama 3 70B), Kunou 72B (Qwen2.5 72B) and Magnum v4 72B (Qwen2.5 72B).
- Mid-size: Cydonia 24B v4.3, with a 65,536-token context, the longest here.
- Small: Rocinante 12B v1.1.
Every model except Cydonia has a 32,768-token context. Several model IDs end in “FP8-Dynamic”, which indicates 8-bit (FP8) quantized weights rather than full precision.
Infermatic also serves stock (unmodified) models, and two more whose base models we couldn’t source. We don’t list those. Search the ones we do on the Models page.
Pricing
Infermatic is flat rate with no per-token billing, so there’s no cost per 1,000 chats. The plans are cumulative: each one includes everything in the plan below it.
- Free, $0: web chat only, no API. 300 requests a day, with responses capped at 60 tokens. Rocinante 12B is the model named for this plan. With 60-token replies, it’s a way to try the interface rather than to roleplay.
- Essential, $9/mo: “curated models up to 72B”, including Magnum v4 SE, Cydonia 24B v4.3 and Anubis 70B v1.1. API access at 1 request at a time, 12 a minute. Web responses up to 2,048 tokens.
- Standard, $16/mo: “all listed models up to 72B”. API at 1 at a time, 15 a minute.
- Plus, $20/mo: all models, including MoE, beta and “72B+” models, with early access. API at 2 at a time, 18 a minute. It also includes monthly credits for premium mainstream models such as DeepSeek V4 Pro. These aren’t part of the flat rate, the credits reset monthly with no rollover, and Infermatic doesn’t state the amount.
Essential is the plan most people will want. Standard is worth it if a model you want isn’t on the curated list. Plus is harder to justify for uncensored use today: nothing larger than 72B is currently listed, although the plan advertises “72B+”.
Is flat rate cheaper? NanoGPT sells Anubis 70B v1.1 at $0.31 / $0.31 per 1M tokens (input / output), about $0.47 per 1,000 short chats. At that rate, you’d need about 19,000 short chats a month before Essential’s $9 comes out ahead. Long roleplay sessions use far more tokens per turn, so heavy users get there much sooner.
Privacy
Infermatic scores 40 / 100 on our privacy rubric, which only gives points for documented claims:
| Rubric item | Infermatic | Points |
|---|---|---|
| Prompt retention | States it doesn’t store input prompts or model outputs | 20 / 20 |
| No training on prompts | Not addressed explicitly | 0 / 15 |
| Technical guarantee (TEE / E2EE) | No: policy only | 0 / 20 |
| Signup | Account required | 3 / 8 |
| Crypto payments | Accepted | 7 / 7 |
| Jurisdiction | Delaware, US | 10 / 10 |
| Upstream hosts disclosed | Not stated | 0 / 10 |
| Independent audit | No | 0 / 10 |
That’s joint 14th of 37, ahead of the other flat-rate roleplay hosts we cover, such as Arli AI (33) and Featherless (23). The no-logging statement is clear. The gap is training: the policy doesn’t say either way. Our rubric only awards those points for an explicit statement.
Access and payment
- OpenAI-compatible API on all paid plans, plus a web chat.
- Account required.
- Card or crypto.
- Infermatic Inc., Delaware, US.
How it compares
- More models: Arli AI is $10/mo for models up to 31B at 16K context, and $15/mo for models up to 355B. We list 131 models there. It scores 33 on privacy.
- Abliterated models, flat rate: Featherless is $25/mo for unlimited tokens at 32K context, and 18 of the 25 models we list there are abliterated. It scores 23.
- More private on paper: AionLabs scores 48 with a no-storage, no-training policy, but bills per token from $0.70 / $1.40.
Verdict
Infermatic is a good default for SillyTavern users who roleplay a lot. Essential at $9 gets you API access to well-known 70B fine-tunes, a clear no-logging policy, crypto payment, and a named company.
The limits are throughput and catalog. One request at a time and 12 a minute is fine for one person chatting, not for anything automated. The catalog is small and topped out at 72B. And if you want models with refusals removed rather than roleplay fine-tunes, look at the abliterated catalogs elsewhere.
See every Infermatic model on its provider page, or compare every provider in the index.