Overview
Novita AI is an inference cloud aimed at developers and startups. Its serverless API covers more than 200 models, with deep coverage of Chinese open-weight families (DeepSeek, Qwen, GLM, Kimi, MiniMax) as well as Llama and Gemma, plus image, video, speech and OCR models. It also rents GPU instances and offers sandboxes for running AI agents.
Prices are low on small and mid-size models: Llama 3.1 8B at $0.02 in and $0.05 out per million tokens, DeepSeek V4 Flash at $0.14 / $0.28 with cheaper cache reads. Batch inference is 50% off input and output on supported models as an introductory offer.
Check each model's page before committing: the pricing page and homepage showed different prices for at least one model on 2026-09-29, and the catalogue changes frequently. For planning, treat Novita as a place to run open models cheaply rather than a source of proprietary models: everything on the price list is an open-weight or third-party media model, billed per token, per image, per video or per character depending on the task.
Pricing and plans
| Plan | Price | What you get |
|---|---|---|
| Llama 3.1 8B Instruct | $0.02 in / $0.05 out per 1M | Serverless |
| DeepSeek V4 Flash | $0.14 in / $0.28 out per 1M | Cache read $0.028; 1M context |
| Qwen3.8 Flash | $0.15 in / $0.47 out per 1M | Serverless |
| DeepSeek V4 Pro 0813 | $1.32 in / $3.96 out per 1M | Cache read $0.132 |
| Kimi K2 Instruct | $0.57 in / $2.30 out per 1M | Serverless |
| Batch inference | 50% off | Introductory discount on supported models |
| Wan 2.6 text-to-video | $0.50 per video | 5 s at 720p |
What it does
- Large open-model catalogue
- 200+ text, image, audio and video models behind one API.
- Batch inference
- Introductory 50% discount on input and output tokens for supported models.
- Cache-read pricing
- Discounted cached input on DeepSeek and other models, e.g. $0.028 per 1M on DeepSeek V4 Flash.
- GPU instances
- Rent GPUs such as 4090, A100 and L40 and pay for what you use.
- Agent sandbox
- Hosted sandboxes for running AI agents alongside model APIs.
Strengths and limitations
Strengths
- Very low prices on small open models.
- Models, GPUs and agent sandboxes under one account.
- Broad coverage of Chinese open-weight model families.
Limitations
- Prices shown differ between pages for some models.
- The batch discount is labelled introductory and may change.
- No free tier or compliance certifications confirmed on the pages read.
Who it suits
Good for
- Cost-sensitive apps on DeepSeek, Qwen or GLM models
- Startups that also need GPUs or media generation
Look elsewhere if
- Regulated workloads that need documented compliance
- Teams wanting proprietary frontier models
Alternatives to Novita AI
| Tool | From | Free option | Stages |
|---|---|---|---|
| Gemini API AI model API (model maker) | Usage-based | Free tier | |
| Scaleway Generative APIs Inference provider (EU-hosted open-weight models) | Usage-based | ||
| SambaNova Cloud (SambaCloud) Inference provider (open-weight models) | Usage-based | Free tier | |
| DeepInfra Inference provider (open-weight models) | Usage-based | ||
| Mistral API (Mistral Studio) AI model API (model maker) | Usage-based | Free tier | |
| Cloudflare Workers AI Serverless inference on an edge network | Usage-based | Free tier |
Questions
How much does DeepSeek V4 Flash cost on Novita AI?
The pricing page listed $0.14 per 1M input tokens ($0.028 cache read) and $0.28 per 1M output tokens on 2026-09-29.
Does Novita AI offer batch discounts?
Yes. Batch inference was available at an introductory 50% discount on input and output tokens for supported models.
Does Novita AI rent GPUs?
Yes. Novita runs a GPU marketplace with cards such as 4090, A100, L40 and 3090, billed for what you use.
Sources read for this page