Cerebras Previews 750-Tokens-per-Second Mode for GPT-5.6 Sol, Speedruns Humanity's Last Exam in 11 Hours
Cerebras claims its new Ultrafast mode runs OpenAI's GPT-5.6 Sol at up to 750 tokens per second — 14× faster than standard — and used it to complete Humanity's Last Exam in 11 hours and 11 minutes.
Cerebras previewed an Ultrafast mode for OpenAI's GPT-5.6 Sol on Thursday, and the numbers it attached are the kind that reframe what's possible in an agent loop. @cerebras claims up to 750 tokens per second — roughly 14 times faster than standard processing — while maintaining comparable accuracy. To dramatize it, the company said it "speedran" Humanity's Last Exam, one of the field's hardest benchmark suites, in 11 hours and 11 minutes.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.