Novita Coupons & Deals
We may earn a commission when you buy through our links. Affiliate disclosure
Coupons and deals
Novita AI: introductory 50% discount on supported Batch inference tokens
Novita advertises an introductory 50% discount on input and output tokens for supported models using Batch inference.
This applies to eligible asynchronous Batch workloads, rather than all real-time model requests, dedicated endpoints, GPUs or Sandbox resources. Check the current supported-model list and rate before submitting work. Batch results are processed asynchronously; the documentation describes a 24-hour completion window and charging for completed requests.
No introductory-offer end date was specified.
Novita AI: eligible new users can receive $100 in Sandbox credits
Eligible new Novita users can complete the account-setup survey to receive $100 in promotional credits for qualifying Sandbox usage. The official documentation says no credit card is required.
The credits apply to eligible Sandbox CPU, RAM and billable storage where applicable. They do not establish a $100 allowance for every Novita model API or GPU product. Availability, validity, quota and usage-scope restrictions can apply; inspect the account voucher status. After exhaustion or expiry, continued usage is billed at the current usage-based rates.
Deal Activated, no coupon code required!
Eligible new Novita users can complete the account-setup survey to receive $100 in promotional credits for qualifying Sandbox usage. The official documentation says no credit card is required.
The credits apply to eligible Sandbox CPU, RAM and billable storage where applicable. They do not establish a $100 allowance for every Novita model API or GPU product. Availability, validity, quota and usage-scope restrictions can apply; inspect the account voucher status. After exhaustion or expiry, continued usage is billed at the current usage-based rates.
Expires:
Submitted: 2 days ago
Choose the service that performs the actual work
Novita’s hosted APIs run supported models for an application, while its GPU products provide compute for workloads the developer operates. Agent Sandbox supplies a separate environment where an agent can run code and use tools. These layers can work together, but purchasing one does not automatically include unlimited use of the others.
A document-processing app may use hosted inference, while a custom model may need dedicated GPU resources.
Online requests versus asynchronous batches
Model pricing distinguishes input, output and any supported cache rates. Image, audio and video operations have their own units and conditions. Estimate the workload using the chosen endpoint rather than applying one advertised language-model rate to every medium.
Batch inference processes multiple requests asynchronously when an immediate answer is unnecessary. The guide lists evaluation, classification and document summarization as example uses. Requests in one batch must target the same model and carry identifiers that let the application match results back to inputs.
The batch has its own processing status and completion window. A successful submission is therefore not the finished answer. Retrieve the output and inspect individual failures before marking an entire processing job complete. Save the results within the documented retention period rather than treating the batch service as a permanent archive.
A Sandbox is a runtime with a lifecycle
Novita Sandbox gives an agent an isolated filesystem and execution environment. Templates define a starting setup, while snapshots preserve a prepared state for reuse. This is useful when several tasks need the same dependencies without rebuilding the environment from scratch each time.
Running, Paused and Killed are distinct states. Pausing preserves filesystem and memory state for resumption; killing terminates the sandbox and cannot be reversed by resuming it. Timeouts and idle-time settings can determine which transition occurs.
Configure lifecycle and durable output storage deliberately for interrupted workflows.
Compute and retained storage have separate costs
Sandbox CPU and allocated memory are billed while running. When paused, those compute charges stop, but retained storage can still be billable beyond the included allowance. Templates and snapshots also occupy persistent storage.
This prevents a misleading cost assumption: paused does not necessarily mean the whole account has zero ongoing usage. Review both runtime duration and saved environments when evaluating a long-running agent project.
GPU instances provide dedicated virtual machines with control over the environment. Serverless GPUs instead handle on-demand execution and scale around submitted work. A predictable sustained workload and an occasional burst can therefore call for different operating and billing choices.
Evaluate the complete application path
The console and documentation expose usage, billing and model-specific behavior. Include input, output, runtime and retained storage in a representative cost estimate.
Current pricing advertises a supported-model batch introduction and eligible Sandbox promotional access under separate terms. Those offers are scoped to their specified products; they do not establish a blanket discount on every GPU or model. Confirm availability, credit validity and the chosen service’s rate before committing a production workload.
An OpenAI-compatible format helps connect supported clients, but capabilities and context limits still vary by endpoint.
Novita Black Friday Deals 2026
Novita AI: eligible new users can receive $100 in Sandbox credits
No coupon code required.
Eligible new Novita users can complete the account-setup survey to receive $100 in promotional credits for qualifying Sandbox usage. The official documentation says no credit card is required.
Novita AI: introductory 50% discount on supported Batch inference tokens
No coupon code required.
Novita advertises an introductory 50% discount on input and output tokens for supported models using Batch inference.
How to Claim Novita Deals
- Open the Novita offer and choose its qualifying plan, billing period or trial.
- Review the included features, usage allowance and total charge. Annual prices require the stated upfront commitment.
- Check the order summary and renewal terms, then complete the purchase or signup when the offer matches your needs.
Novita FAQs
Is Agent Sandbox the same product as a hosted model API?
No. The model API supplies inference. Sandbox provides a runtime for code, files and tool use; a workflow may use both with separate usage costs.
Does pausing a Sandbox stop every charge?
It stops CPU and RAM billing, but retained persistent storage can still incur charges beyond the included allowance. Templates and snapshots also use storage.
Can a killed Sandbox simply be resumed?
No. The documentation distinguishes a resumable paused environment from a killed one whose runtime resources are released. Configure timeout behavior accordingly.
Can one batch mix requests for different models?
The current Batch guide says every request within one input file must use the same model. Track each request’s identifier and inspect output or error results after processing.
Loading comments....