Cerebras runs open models on wafer-scale hardware, and like Groq its pitch is throughput rather than a proprietary frontier model: output tokens arrive fast enough to change what a voice agent or an interactive tool feels like. Billing is ordinary per-token pricing, so the interesting question is whether the speed is worth the per-token rate for your workload.
| Model | Input $/1M | Output $/1M | Best for |
|---|---|---|---|
| GPT-OSS-120B | $0.35 | $0.75 | Fast general-purpose open model |
| ZAI-GLM-4.7 | $2.25 | $2.75 | Higher-capability tier |
| Catalogue range | $0.50–$1.50 | varies | Most hosted models sit in this band |
→ Estimate your bill on the API cost calculator or model a whole app with the AI app cost estimator.
Yes. Cerebras offers a free developer tier with rate limits suitable for building and evaluating, with no card up front. The self-serve Developer tier starts at about $10 and raises rate limits roughly tenfold while giving your requests higher scheduling priority. If a free quota is your main filter, compare against Groq, Gemini and Mistral on the free API tiers page.
1. Sign up at cloud.cerebras.ai.
2. Open API Keys and create a key — copy it once, it is not shown again.
3. Start on the free tier; add the $10 Developer tier when rate limits become the bottleneck.
4. The endpoint is OpenAI-compatible, so existing OpenAI SDK code works by changing the base URL and key.
Test it:
If raw speed is not the requirement, almost anything is cheaper per token: DeepSeek and Groq's small Llama models are the usual budget picks, and OpenRouter gives you one key across providers so you can A/B the same prompt on several. Where latency genuinely drives revenue — voice agents, interactive coding tools — model the cost of the slower option properly with the LLM latency calculator before assuming the cheap provider wins.
Yes. Cerebras Inference has a free developer tier with rate limits that is enough to build and test against, no card required. Moving to the self-serve Developer tier costs about $10 and unlocks roughly 10x higher rate limits plus higher-priority processing.
Reference pricing (July 2026): GPT-OSS-120B is around $0.35 input / $0.75 output per million tokens, and ZAI-GLM-4.7 around $2.25 / $2.75. Across the catalogue most models fall in a $0.50-$1.50 per million range, billed per input and output token against a published rate card like any other provider.
Create an account at cloud.cerebras.ai, open the API Keys section, generate a key and copy it once. The free tier is active immediately; add the $10 Developer tier when your rate limits start biting.
Both sell speed on open models and both are far faster than commodity GPU inference. Compare on the specific model you need — the catalogues differ, and a model available on one may not be on the other. Where both host the same model, compare tokens per second alongside the per-million rate rather than price alone.
Not affiliated with Cerebras. Prices are reference estimates — always verify on the official pricing page.