Cerebras
6/10Cerebras runs a hosted inference cloud that serves open-weight AI models through an API on its own wafer-scale chips, aimed at developers and enterprises that need low latency.
Data last updated November 2026 · Our methodology
Overview
Cerebras Inference is a hosted API that serves open-weight language models on the company's wafer-scale processors (WSE-3 and CS-3 systems, with the CS-4 launched in August 2026). Independent benchmarks back its main selling point, which is generation speed. Artificial Analysis ranked it the fastest of 18 providers for gpt-oss-120b, at roughly 1,600 to 1,700 tokens per second. That makes it a good fit for voice agents, interactive coding assistants and multi-step agent loops where latency matters. The API is OpenAI-compatible, and Cerebras publishes official Python and Node.js SDKs. Cerebras is also offered through AWS Marketplace, OpenRouter, Hugging Face and Vercel. Pricing is per million tokens. New users get a one-time $5 credit that expires after 30 days. The self-serve developer tier is pay-as-you-go: gpt-oss-120b costs $0.35 per million input tokens and $0.75 per million output tokens. Cerebras Code subscriptions were reported at $50 and $200 per month but were listed as sold out. Enterprise contracts are quoted individually and add higher rate limits, priority and fine-tuning. Customers in regulated industries can also install CS-3 hardware in their own data centers. The main limitation is the very small public model catalog. Recent reviews counted only two models on the public endpoint after several were deprecated, and developers have complained that rate limits kick in quickly. This is not general-purpose hosting: there are no virtual machines, storage or databases.
Our Rating
AI Software5.9/10
Hosting & Infrastructure6.1/10
Who Cerebras is best for
Best software for DevelopersGood fit
It offers a self-serve, OpenAI-compatible API and official SDKs with pay-as-you-go token pricing and very fast generation for coding and agent workloads.
Enterprise agreements add priority access, higher limits and fine-tuning, and CS-3 systems can be installed on-premise.
Low-latency inference suits AI features built into products, though the small model catalog limits choice.
Less suited
On-premise CS-3 deployment is presented as an option for regulated and government workloads.
Pros & Cons
Pros
- Independent benchmarks rank it the fastest provider for gpt-oss-120b, at about 1,600 to 1,700 tokens per second.
- The API is OpenAI-compatible and has official Python and Node.js SDKs, so teams can switch existing code over with little rework.
- Token prices for gpt-oss-120b are low, at $0.35 per million input and $0.75 per million output tokens.
- It is available through AWS Marketplace, OpenRouter, Hugging Face and Vercel as well as directly.
- Holds SOC 2 Type II, and CS-3 hardware can be deployed on-premise.
Cons
- The public model catalog is very small; recent reviews counted only two models after several were deprecated.
- Users report that rate limits are hit quickly, and the old free daily-token tier has been replaced by a one-time $5 credit.
- Fine-tuning is offered only on enterprise contracts.
- HIPAA is not listed on the trust center, so healthcare buyers need to confirm a BAA with sales.
- Cerebras Code subscriptions were reported as sold out.
Key Features
Not included
- Managed WordPress
- Object Storage
- Shared Hosting
Cerebras Pricing
Trial credit
- For evaluation, not production
- Credit expires 30 days after activation
- One-time $5 credit after adding a payment method
- Access pauses when the credit runs out unless you buy pay-as-you-go credits
Developer (pay-as-you-go)
- Self-serve token billing
- 131K context reported by a third-party review
- $10 minimum deposit reported by a third-party review
- Higher rate limits and queue priority than the trial tier
Enterprise
- All supported models
- Available by agreement
- Enhanced latency options
- Higher priority and rate limits
- Model-weight customization and fine-tuning
Frequently Asked Questions
New users who add a payment method get a one-time $5 credit that expires after 30 days. You are not charged automatically when it runs out; access pauses until you buy pay-as-you-go credits.
Usage is billed per million tokens, with separate input and output prices. For example, gpt-oss-120b costs $0.35 per million input tokens and $0.75 per million output tokens.
The public endpoint has a small catalog. Recent reviews list gpt-oss-120b and one Gemma or Qwen model. Enterprise customers can access all supported models.
Cerebras publishes official SDKs for Python (cerebras_cloud_sdk) and Node.js, plus a REST chat completions endpoint you can call with cURL.
Third-party reviews of its trust center confirm SOC 2 Type II, GDPR and CCPA. HIPAA is not listed, so healthcare buyers should confirm a BAA with sales.
Yes. Cerebras Inference is listed on AWS Marketplace and is also available through OpenRouter, Hugging Face and Vercel.
Not through the self-serve API. Model-weight customization and fine-tuning are part of enterprise agreements.
Similar Tools
See all alternatives
Amazon's public cloud platform, with over 200 pay-as-you-go services for compute, storage, databases, networking and AI, used by startups through large enterprises and governments.
Portkey is an AI gateway and LLMOps control plane that routes, monitors, and governs requests to 1,600+ LLMs for engineering teams and enterprises.
Northflank is a Kubernetes-based developer platform for deploying services, jobs, databases and GPU workloads on its managed cloud or on customers' own cloud accounts.
Ranked near Cerebras in AI Software
See Cerebras in the listCerebras is #569 in AI Software by our score.
