Novita
Verified Oct 16, 2026 ActivePay-as-you-go APIs for 200-plus open models, rented GPUs, and a sandbox for running agents.
Listed on EffectHub ·
Pricing
- Pricing model
- Paid
Metered, not a plan. Model APIs bill per million tokens — published rates run from about $0.03 per million input tokens for small models to a few dollars for frontier open models, with a 50% introductory batch discount — and GPUs bill per second. No unit in this catalog matches per-token billing, so no entry figure is recorded; fees are non-refundable except for billing errors.
Use tool ↗About Novita
Novita AI is an inference and compute cloud with three parts: serverless endpoints for more than 200 text, image, audio and video models billed per token; dedicated GPU instances and bare-metal clusters billed per second; and an agent sandbox that runs agent code in isolation. Rates are published openly per model — a small open model runs about 14 cents per million input tokens, while a frontier open model runs a few dollars — and batch inference carries an introductory discount. The terms name Delaware law and Wilmington courts, and the company states a zero data retention policy. Hugging Face and Poe are listed as integration points.
- Platforms
- Web · API
- API
- Yes
- Last verified
- Oct 16, 2026
Verified sources
- Pricing page lists per-model rates, e.g. DeepSeek V4.1 Flash $0.3 input / $1.2 output per million tokens and GLM 4.7 Flash $0.07 input, plus a 50% introductory batch-inference discount (15 October 2026)novita.ai ↗
- Terms of service: Delaware governing law (section 20), zero data retention policy (10.2), all fees non-refundable including unused credits (section 8) (15 October 2026)novita.ai ↗
Is Novita legit?
We don’t label tools safe or unsafe. Below are the records we checked ourselves, each with the date we read it.
- Domain registered
- Sep 18, 2023 (3 years ago)Registry record (RDAP), registrar GoDaddy.com, LLC, looked up Sep 12, 2026
- TLS certificate
- Issued by Let's Encrypt, valid until Dec 22, 2026Seen in our own connection to the site on Oct 16, 2026
- Last checked
- Oct 16, 2026The site answered
- Operated by
- Novita AI, United StatesOur research, card verified Oct 15, 2026
Pros & cons of Novita
Novita covers the whole ladder — call someone else's open model by the token, rent the GPU, or run an agent in a sandbox — which spares you stitching three vendors together. Published per-model rates make it easy to compare against Together or Fireworks. The trade is that everything is metered, so cost control is your job, and refunds do not exist.
Pros
- More than 200 models across text, image, audio and video behind one API
- Per-model rates published openly, down to a few cents per million input tokens
- GPU instances and bare-metal clusters billed by the second
- Agent sandbox for isolated execution alongside the model APIs
- Stated zero data retention policy and an introductory batch-inference discount
Cons
- Everything is metered — there is no flat plan to cap spend
- All fees are non-refundable per the terms, unused credits included
- Separate cache-read and batch rates make like-for-like comparison with other providers fiddly
- Only a Delaware governing-law clause identifies the jurisdiction; no address is published
Best for
- Cheap inference against open models like the DeepSeek family
- Renting GPUs by the second instead of by the month
- Running untrusted agent code in an isolated sandbox
Not for
- Teams that need a predictable monthly bill
- Workloads requiring closed frontier models from their original vendors
Watch out: Cache-read rates are quoted separately from input rates, so a naive per-token comparison against another provider can be off by a lot in either direction.