Cerebras ditches plan cards for per-model token pricing starting at $0.35/M
By Rob Litterst, Co-founder, PricingSaaS
Published
Cerebras’s pricing page · Change detected
What changed
Cerebras retired its three tiered plan cards and replaced them with a two-column Developer/Enterprise comparison table. The page now discloses per-model API rates directly: GPT OSS 120B at $0.35 per million input tokens and $0.75 per million output tokens, and Qwen 3.8 27B at $0.99/$1.49 per million tokens.
Why it matters
For an inference provider whose core pitch is raw speed, abandoning packaged tiers for transparent per-token rates is a bet that price transparency, not plan structure, is what wins developer trust in a crowded GPU-inference market. Watch whether Cerebras keeps this table stable for more than a few weeks, or whether the same 10-day cadence produces another restructuring before it settles.
In context
Cerebras has touched credits and usage billing three times in under two months — renaming Free to Free Trial with $5 credit framing in July renaming Free to Free Trial, then revealing a per-model rate table a per-model rate table, and now retiring its plan cards for a straight Developer/Enterprise comparison. That's the fastest pace in the index: seven real changes in under two months, this one landing five days after Cerebras rewrote its Free Trial pitch rewrote its Free Trial pitch. It joins roughly 330 companies — including HubSpot and Datadog — that shifted pricing toward usage-based credits in the last six months, though few touch the pattern this often. For a speed-focused inference provider, ditching tiers for raw per-token rates bets that transparency, not packaging, is now the differentiator.
What people are saying
“@cerebras is fast, but I burned through $5 USD in about 6 minutes, adding a small feature to an already written web app. 6minutes. (no it didn't finish the task) - i'm not convinced the pricing is quite right for the output. neat toy though if you wanna burn $$”
View original“I have always wanted to test out Cerebras for speed. So I signed up with their cloud on a paid plan. The demo showed 2,016 tokens/sec with Qwen 3.8 27B. Hermes reported 44. That's not a controlled comparison, but I kept hitting rate limits and waiting a minute to continue. I'm disappointed with the experience so far. What am I doing wrong here?”
View original“Qwen3.8-27B 上了 Cerebras，速度拉满：约 1,850 tokens/s，输入 $0.99/M、输出 $1.49/M。64k 上下文免费，128k 收费，Artificial Analysis 智能指数给到 34。 1,850 tok/s 是什么概念 —— 写代码补全几乎无感延迟，跑 agent 循环能快一轮是一轮。 开源模型的竞争早就从 "谁更聪明" 卷到 "谁跑得快还便宜" 了。 #Qwen #Cerebras #AI #vibecoding”
View original