Provider pricing
Cerebras AI API pricing
Cerebras Inference runs on the WSE-3 wafer-scale chip, currently the fastest published inference platform — 2000+ tokens/sec on Llama 4 Scout and 1000+ on Llama 4 Maverick. Pricing is per-token with a free tier, and the API is OpenAI-compatible for drop-in replacement.
Founded2016
HeadquartersSunnyvale, CA
Tracked models3
Best for
Workloads where raw throughput is the bottleneck — agents, batch generation, long-context summarization.
Strengths
- Fastest published inference: 2000+ tok/s on Llama 4 Scout
- Wafer-scale memory keeps full models on a single chip — no sharding latency
- OpenAI-compatible API with streaming, tool use, JSON mode
- Free developer tier for prototyping and benchmarking
Tracked Cerebras models
| Model | Input $/M | Output $/M | Context | Tier | Best for |
|---|---|---|---|---|---|
| Llama 3.3 70B (Cerebras) | $0.850 | $1.20 | 128,000 | Mid-tier | Wafer-scale Llama 3.3 70B — 2000+ tokens/sec inference |
| Llama 3.1 8B (Cerebras) | $0.100 | $0.100 | 128,000 | Budget | Wafer-scale 8B for ultra-low-latency chat |
| Qwen3 32B (Cerebras) | $0.400 | $0.800 | 128,000 | Mid-tier | Wafer-scale Qwen3 32B at 2000+ tokens/sec inference |