Cerebras announces the $50/month Code Pro and the $200/month Code Max plans, offering users access to Qwen3-Coder at speeds of up to 2,000 tokens per second
Two interesting examples of inference speed as a flagship feature of LLM services today. Bluesky: Tim Kellogg / @timkellogg.me : Cerebras Code — use models hosted on Cerebras with a dev-friendly subscription model — this is sick, but missed opportunity by not making it compatible with Anthropic API / Claude Code — www.cerebras.ai/blog/introdu... X: @cerebrassystems : Cerebras Code: 20x faster than Claude, 1x the price Today we are launching two monthly coding plans: ➡️Cerebras Code Pro: $50/m - for indie developers ➡️Cerebras Code Max: $200/m - for power users with 5x rate limits Both plans get: Qwen3-Coder at 2,000 tokens/s, 131K context, [video] Anton / @abacaj : Why pay for this if the cost is the same and you get access to a worse model? I get the speed factor but for code - I want the best model Petri Kuittinen / @kuittinenpetri : Qwen3-Coder-480B-A35B on Cerebras is insanely fast & good. 2000+ tokens/s. I got this slick web page including images, 748 lines of code generated in 1.5 seconds. Full prompt used: Generate home page for an AI company. Invent a slogan and content for it. Include all the usual [image] Simon Willison / @simonw : Cerebras offering their lightning fast (2,000 token/second) hosted version of Qwen3-Coder-480B-A35B-Instruct - a model that was released just ten days ago on July 22nd - for $50/month or $200/month Yam Peleg / @yampeleg : This is Qwen3-Coder-480B-A35B. (not the 30B) Roughly ~same performance as Claude Sonnet. 6x the speed. Speechless. George Cameron / @grmcameron : Cerebras Code could be really promising for power-users who appreciate Cerebras' speed and are cost conscious If avg. workload is 5k input & 1k output (sensitive estimate, potential for a lot more input heavy), one would save $130/month when subscribing to the @CerebrasSystems [image] Forums: Hacker News : Cerebras Code
Context & Ripple Effects
Cerebras is turning the speed positioning of its earlier AI inference service into a developer-facing coding subscription. The offer packages hosted access to Qwen3-Coder shortly after Alibaba introduced the model and its accompanying coding CLI, expanding the options for agentic coding workflows.
The pricing also arrives as premium AI tiers increasingly distinguish users by access and capacity. Anthropic’s planned limits for heavy Claude users, aimed at curbing continuous background use, illustrate why a higher-priced plan with five times the rate limits matters.
First-order effects
- Developers can buy hosted Qwen3-Coder access through $50 Pro and $200 Max plans, with Cerebras making throughput—up to 2,000 tokens per second—a central product attribute.
- Cerebras segments coding demand by capacity: Max buyers receive five times the rate limits, while the company positions the service against Claude on speed and price.
Second-order effects
- Coding-model providers and hosted inference services face sharper comparison on interactive latency and usable rate limits, not just model quality or nominal subscription price.
- Higher-capacity subscribers may shift intensive coding workloads toward services that can sustain faster output, making rate-limit policy a more visible competitive lever.
Third-order effects
- If this packaging holds, coding AI subscriptions will increasingly sell differentiated inference capacity—speed, context and limits—as much as model access, linking product tiers directly to inference infrastructure advantages.
- The model layer may become more separable from the service layer: open coding models can be competitively differentiated by the host’s delivery performance and capacity economics rather than exclusivity alone.
The trend: AI coding services are evolving from model-centric subscriptions into capacity-tiered products where inference speed and rate limits define the premium experience.