How much does Qwen3.8-27B cost on Cerebras?

Cerebras lists Qwen3.8-27B at $0.99 per million input tokens and $1.49 per million output tokens. A free trial tier costs nothing and allows 5 requests per minute, 30,000 uncached tokens per minute and 1 million tokens per day, which is enough to prototype an agent before moving to a paid Developer plan.

How does its speed compare to other models on Cerebras?

Cerebras documents Qwen3.8-27B at roughly 1,500 tokens per second. The other model in its public catalog, OpenAI's 120B gpt-oss-120b, is listed at roughly 3,000 tokens per second. Cerebras says many additional model families are reachable through Dedicated Endpoints rather than the public catalog.

What context window do I actually get?

Qwen3.8-27B on Cerebras gives a 64K-token context and 32K maximum output on the free trial, rising to a 128K-token context and 40K maximum output on paid tiers. Cerebras also caps the free trial at 30,000 uncached tokens per minute, so long-context work in a loop will hit the rate limit before it hits the context limit.

Can Qwen3.8-27B read images through this endpoint?

Yes. Cerebras documents Qwen3.8-27B as a multimodal endpoint that accepts text and image inputs in PNG and JPEG form. It also supports reasoning, tool calling, streaming responses and structured outputs, so it can drop into an existing agent stack without giving up function calling or JSON-shaped replies.

Cerebras · 2026-09-03 · major

Qwen3.8-27B on Cerebras — 1,500 tokens per second at $0.99 per million

Cerebras now serves Qwen3.8-27B on its public endpoints at about 1,500 tokens per second, with a 128K context on paid tiers and pricing of $0.99 per million input tokens and $1.49 per million output tokens.

Cerebras

Cerebras added Qwen3.8-27B to its public endpoints, running Alibaba's 27B dense multimodal model at roughly 1,500 tokens per second.

Key specs

Speed	~1,500 tokens/sec

Quick facts

Model id	qwen-3.8-27b
Maker	Alibaba (Qwen); served by Cerebras
Context window	64K free trial / 128K paid
Max output	32K free trial / 40K paid
Inputs	Text and images (PNG, JPEG)
Features	Reasoning, tool calling, streaming, structured outputs
Rate limits	5 req/min free; 300 req/min on Developer

Pricing

Input	$0.99 / 1M tokens
Output	$1.49 / 1M tokens
Free Trial · 5 req/min, 1M tokens per day	$0

source ↗

What is it?

Model id `qwen-3.8-27b` is now listed in the Cerebras model catalog, meaning Alibaba's 27B dense multimodal model runs on Cerebras wafer-scale hardware through the public API rather than a dedicated endpoint. Cerebras documents it at about 1,500 tokens per second. It accepts text plus PNG and JPEG images, and supports reasoning, tool calling, streaming and structured outputs.

How does it work?

Access splits into two documented tiers. The free trial gives a 64K context, 32K maximum output, 5 requests per minute, 30K uncached tokens per minute and 1 million tokens per day. The Developer tier raises that to a 128K context, 40K maximum output, 300 requests per minute and 150K uncached tokens per minute with no daily cap. Cerebras notes that model families outside this public catalog are reachable only through Dedicated Endpoints.

Why does it matter?

Speed changes what you can build with Qwen3.8-27B, not just how long you wait: at roughly 1,500 tokens per second, agent loops and multi-step tool chains that used to stall on generation finish while a user is still watching. The published price of $0.99 per million input tokens also puts a fast open-weight model in reach for interactive products, and the free trial is enough to test an agent before committing.

Who is it for?

developers building latency-sensitive agents and coding tools

Frequently asked questions

How much does Qwen3.8-27B cost on Cerebras?: Cerebras lists Qwen3.8-27B at $0.99 per million input tokens and $1.49 per million output tokens. A free trial tier costs nothing and allows 5 requests per minute, 30,000 uncached tokens per minute and 1 million tokens per day, which is enough to prototype an agent before moving to a paid Developer plan.
How does its speed compare to other models on Cerebras?: Cerebras documents Qwen3.8-27B at roughly 1,500 tokens per second. The other model in its public catalog, OpenAI's 120B gpt-oss-120b, is listed at roughly 3,000 tokens per second. Cerebras says many additional model families are reachable through Dedicated Endpoints rather than the public catalog.
What context window do I actually get?: Qwen3.8-27B on Cerebras gives a 64K-token context and 32K maximum output on the free trial, rising to a 128K-token context and 40K maximum output on paid tiers. Cerebras also caps the free trial at 30,000 uncached tokens per minute, so long-context work in a loop will hit the rate limit before it hits the context limit.
Can Qwen3.8-27B read images through this endpoint?: Yes. Cerebras documents Qwen3.8-27B as a multimodal endpoint that accepts text and image inputs in PNG and JPEG form. It also supports reasoning, tool calling, streaming responses and structured outputs, so it can drop into an existing agent stack without giving up function calling or JSON-shaped replies.

Try it

Call the Cerebras API with model id `qwen-3.8-27b`