AIFind
Novita AI
AI cloud for developers: 200+ models through one API, agent sandboxes and GPU instances, billed by usage
Novita AI is a developer-focused AI cloud that puts model APIs, agent sandboxes and GPU infrastructure on one platform. Its serverless model APIs expose 200+ models for text, image, audio, video and vision through a single OpenAI-compatible API, billed per token. Agent Sandbox gives agents isolated runtimes that start in about 200 ms and bill per second, and the GPU side offers dedicated GPU instances, serverless GPUs that scale to zero, and bare-metal clusters on NVIDIA H100 and H200 hardware. Everything is pay as you go, with no flat subscription plans, and the company says it costs up to 50% less than major cloud providers.
- Best for
- Developers, startups and AI product teams that want to call many open models through one API, run agents in isolated sandboxes, or rent GPUs without a long-term commitment
- Pricing
- Pay as you go: model APIs billed per million tokens (for example DeepSeek V4.1 Flash at $0.30 input and $1.20 output), sandboxes billed per second, GPUs priced by type; new users can claim $100 in promotional sandbox credits with no card
Pros
- One OpenAI-compatible API gives access to 200+ models, including DeepSeek, GLM, Kimi, Qwen and Llama families
- Pure usage-based billing with no subscription: tokens, seconds or GPU time
- Agent sandboxes start in about 200 ms, bill per second and stop billing when paused
- Covers the full stack from model APIs to serverless GPUs and bare-metal H100 and H200 clusters
- New users can claim $100 in promotional sandbox credits with no credit card
Cons
- Built for developers: there is no consumer chat app, so you need an API key and code to use it
- Usage-based bills can be hard to predict for variable workloads, and the promotional credits apply to sandbox usage rather than all services
- Some models use tiered pricing and GPU prices depend on the instance type, so you have to check the pricing page for your exact setup
- The savings claim of up to 50% against major clouds comes from the company and depends on your workload
Core features
- • Serverless model APIs for LLM, image, audio, video and vision models
- • OpenAI-compatible endpoint with Bearer API keys
- • Dedicated endpoints with isolated resources
- • Agent Sandbox with SDK, CLI, snapshots, SSH and computer use
- • GPU instances, serverless GPUs and bare-metal clusters
- • Per-token, per-second and per-execution billing
- • Batch inference with an introductory 50% discount on supported models
- • Cache-read pricing for supported models
- • Documentation available as llms.txt and Markdown for agents
Use cases
- 1. Call open models such as DeepSeek, GLM or Qwen through one OpenAI-compatible API
- 2. Run coding or research agents in isolated sandboxes billed per second
- 3. Deploy a model on a dedicated GPU instance or endpoint
- 4. Run bursty inference jobs on serverless GPUs that scale to zero
- 5. Train or serve large models on bare-metal H100 or H200 clusters
How to use
Create an account and generate an API key. For model APIs, point an OpenAI-compatible client at https://api.novita.ai/openai with your key as a Bearer token and pick a model from the catalog. For agents, install the Sandbox SDK or CLI and start a sandbox, which bills per second while it runs. For heavier work, launch a GPU instance or serverless GPU job and watch usage and balance from the billing page.
Pricing plans
Serverless model APIs
Pay per token, from $0.02 per million input tokens
- • 200+ models for text, image, audio, video and vision
- • OpenAI-compatible API at api.novita.ai/openai
- • Examples: DeepSeek V4.1 Flash $0.30 input and $1.20 output per million tokens, GLM 5.3 Flash $0.15 and $0.50, Kimi K3 $3 and $15
- • Cache-read discounts on supported models
- • Introductory 50% discount on batch inference for supported models
Agent Sandbox
Pay per second for CPU and RAM while running
- • Isolated runtimes with about 200 ms startup
- • 1 vCPU at $0.0000098 per second
- • RAM at $0.0000032 per GiB per second
- • 20 GB ephemeral storage per sandbox and 60 GB free persistent storage
- • $100 promotional credits for new users, no card required
GPU Cloud
Priced by GPU type; see the pricing page
- • Dedicated GPU instances with full control
- • Serverless GPUs that scale to zero and bill for execution only
- • Bare-metal clusters on NVIDIA H100 and H200
- • Dedicated endpoints with isolated resources
FAQ
What is Novita AI?
Novita AI is a developer-focused AI cloud that offers serverless model APIs, agent sandboxes and GPU infrastructure on one platform, all billed by usage.
How does Novita AI pricing work?
It is pay as you go. Model APIs are billed per million tokens, for example DeepSeek V4.1 Flash at $0.30 input and $1.20 output. Sandboxes are billed per second for CPU and RAM, and GPU prices depend on the instance type.
Does Novita AI have free credits?
New users can complete a short account setup survey to receive $100 in promotional credits for eligible Agent Sandbox usage, and no credit card is required. Credits can have limits on validity and scope.
Is the Novita AI API compatible with OpenAI?
Yes. The LLM APIs are OpenAI-compatible at https://api.novita.ai/openai and use a Bearer API key, so existing OpenAI client code can usually point at it.
What hardware does Novita AI offer?
It offers dedicated GPU instances, serverless GPUs that scale to zero, and bare-metal clusters. The site shows 8x NVIDIA H100 and 8x NVIDIA H200 nodes with NVLink and 400 Gb/s RDMA networking.
Novita AI
This link may be an affiliate link — see our disclosure.