3 ms·
> Qwen3-235B uses an efficient mixture-of-experts architecture that delivers exceptional compute efficiency, enabling Cerebras to offer the model at $0.60 per m
by westurner 1y ago
> Qwen3-235B uses an efficient mixture-of-experts architecture that delivers exceptional compute efficiency, enabling Cerebras to offer the model at $0.60 per million input tokens and $1.20 per million output tokens—less than one-tenth the cost of comparable closed-source models.
$ 0.60/million input tokens
$ 1.20/million output tokens
How many minutes of 4K YouTube HDR video is that equivalent to in kWh of energy usage?
> Concurrent with this launch, Cerebras has quadrupled its context length support from 32K to 131K tokens—the maximum supported by Qwen3-235B.