Cerebras

What it is for

Cerebras Instant AI is a signed-out inference lobby: the Instant AI Chat card (see the cover: Chat, “What can Cerebras inference do?”, streamed reply, ~1,283 tokens/sec) is the signed-out playground. Use it when the job is hit a very fast hosted model, not when you want a local runner.

It is not open source. Playground / API start with a one-time $5 promotional credit. Production is pay-as-you-go per 1M tokens.

What makes it unique: the cover is Cerebras’s own Instant AI Chat card — a real prompt and tokens/sec reply — not the wafer poster.

Key features

• Instant AI Chat — prompt, streamed reply, and tokens/sec on the signed-out playground (cover).
• Coding agent card — GLM-4.7 starter snippet on the same lobby.
• Get Free API Key — header CTA on the signed-out page.
• Public token table — GPT-OSS 120B and Qwen 27B dollars on cerebras.ai/pricing.
• $5 promo credit — add a payment method; expires in 30 days; no auto-enroll (pricing FAQ).

Who it's for

Best fit: developers who will burn a $5 credit to measure tokens/sec, then pay per 1M.

Poor fit: people who want a local runtime (Ollama) or a multi-vendor router.

Pricing

Verified 2026-10-21 (Asia/Shanghai) from cerebras.ai/pricing. Re-check before you buy.

• Promo credit — $5 one-time after a valid payment method. Expires 30 days. API and Playground pause when it is gone unless you buy PayGo credits.
• GPT-OSS 120B — $0.35 / 1M input, $0.75 / 1M output (~3000 tokens/s on the table).
• Qwen 27B — $0.99 / 1M input, $1.49 / 1M output (~1,850 tokens/s on the table).
• Enterprise / agreement rows — contact / available by agreement. Do not invent a seat.

The page says features, models, and capacity are subject to availability. We only copied rows that printed $.

Limitations / costs

• $5 is a coupon, not a plan. It expires in 30 days.
• Only two public token pairs were in the payload we copied. Other models may be “contact.”
• The lobby is not the API. Instant AI cards still send you through a key.
• Not local. Weights stay on Cerebras.

Alternatives

• Together AI — hosted inference with public token cards.
• Ollama — local runner; MIT; no invented Cloud dollar.

Official tutorial

Official Cerebras channel video (linked from cerebras.ai): Cerebras Inference in 30 seconds (oEmbed author: Cerebras).

Written cerebras.ai/pricing plus the signed-out Instant AI lobby.

Website

https://inference.cerebras.ai

Source

Catalog row curated for AI Tools Nav. Cover is the Instant AI Chat card cropped from inference.cerebras.ai (prompt + tokens/sec reply), not the full marketing lobby. Dollars checked on cerebras.ai/pricing ($5 promo; GPT-OSS 120B $0.35 / $0.75 per 1M; Qwen 27B $0.99 / $1.49 per 1M). Tutorial: Cerebras Inference in 30 seconds on the official Cerebras channel (oEmbed).

Last verified

2026-10-21 (Asia/Shanghai) editorial check. Only token rows that printed $ were copied.