Cerebras Inference
Open-weight models at thousands of tokens per second on wafer-scale chips
Overview
Cerebras Inference is the cloud API from Cerebras Systems that runs open-weight models on its wafer-scale processors. Self-serve users get GPT OSS 120B and Qwen 3.8 27B through an OpenAI-compatible API, billed per token, while Dedicated Inference adds reserved capacity, more model families and custom fine-tuned weights.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- GPT OSS 120B: $0.35 input / $0.75 output per M tokens, ~3000 tokens/s
- Qwen 3.8 27B: $0.99 input / $1.49 output per M tokens, ~1,850 tokens/s
- Community Discord support
- Production-ready capacity
- Best priority, higher rate limits and enhanced latency options
- Custom model weights, fine-tuning and training services by agreement
- Dedicated account team
What you can produce with Cerebras Inference
- Chat completions from GPT OSS 120B at about 3,000 tokens per second (vendor figure)
- Multimodal text-and-image requests to Qwen 3.8 27B with configurable reasoning
- An OpenAI-compatible API with Python and Node.js SDKs, tool calling, structured outputs and prompt caching
- Reserved Dedicated Inference endpoints running Llama, Mistral, MiniMax, Gemma or Qwen variants
- Fine-tuned custom weights deployed on Cerebras hardware through Dedicated Inference
- Access through AWS Marketplace, OpenRouter, Hugging Face or Vercel as well as direct
Speed is the reason to use Cerebras, and the vendor's figures of roughly 3,000 tokens per second on GPT OSS 120B make tool-calling agents and live chat feel immediate. The API follows OpenAI's format, so switching is mostly a base-URL change, and $0.35 per million input tokens on GPT OSS 120B is not expensive.
The catch is choice. The self-serve menu has shrunk to two models since Llama 3.3 70B was retired in February 2026, and anything broader, including your own fine-tuned weights, means Dedicated Inference and a quote. Use it when latency is the bottleneck and one of its two models fits; look elsewhere if you need a specific model off the shelf.
What we found
Cerebras Inference is the cloud API that Cerebras Systems, the Sunnyvale maker of wafer-scale AI chips, launched in August 2024. The pitch has not changed: run open-weight models on a processor the size of a silicon wafer and return tokens far faster than GPU clouds. Speed figures on the pricing page are Cerebras's own, about 3,000 tokens per second on GPT OSS 120B and about 1,850 on Qwen 3.8 27B, and the site footnotes them as third-party or internal benchmarks that vary by workload.
What has changed is the menu. Earlier versions of this page led with Llama 3.3 70B at 1,800 tokens per second; Cerebras deprecated that model on its shared service in February 2026, along with Qwen 3 32B. Self-serve users now choose between two models, GPT OSS 120B at $0.35 per million input tokens and $0.75 output, and Qwen 3.8 27B, a multimodal model added in September 2026, at $0.99 and $1.49. A broader catalogue, including Llama, Mistral, MiniMax and Gemma variants, lives on Dedicated Inference, which reserves capacity and accepts fine-tuned custom weights. That reverses an older complaint here that custom weights were not supported.
The free tier is also smaller than this page once said. Adding a payment method earns a one-time $5 credit that expires after 30 days; after that, usage is pay-as-you-go, and the trial carries a one-million-token daily cap.
Cerebras listed on Nasdaq as CBRS in May 2026, raising $6.4 billion gross. Its August results reported cloud revenue up 281% year on year and named OpenAI, Cognition, Lovable, Block, Figma and CrowdStrike among customers and partners. OpenAI's Ultrafast mode for GPT-5.6 Sol runs on Cerebras, though that model is sold through OpenAI's API, not this one. The trade-off for developers is plain: very fast tokens on a short list of models, or a sales conversation and a dedicated contract for anything else.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Cerebras Inference's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Cerebras Inference unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Cerebras Inference directly →
Spotted something out of date? Suggest an update →
Alternatives to Cerebras Inference
More in LLMs & Chat
