August 13, 2026
Speed kills, and comments roast
Accelerating GPT-5.6 Sol Ultrafast
OpenAI’s new turbo AI has fans screaming, skeptics side-eyeing, and everyone asking the price
TLDR: OpenAI and Cerebras say their new Ultrafast mode makes GPT-5.6 Sol dramatically quicker while keeping the same quality, potentially changing time-sensitive work. Commenters loved the speed boost, joked about giant “dinner plate” chips, and immediately started arguing over the one missing detail: the price.
OpenAI and Cerebras just teased Ultrafast Mode, a new premium setting for GPT-5.6 Sol that promises answers at eye-watering speed — up to 750 tokens a second, which in normal-person terms means the chatbot can spit out text much faster without supposedly getting dumber. The companies are pitching it as a big deal for jobs where every second counts, like legal work, finance, engineering, outages, and cybersecurity. They also bragged that it chewed through a brutal expert-level test called Humanity’s Last Exam in 11 hours, while a rival model took more than three days.
But the real action was in the comments, where people instantly turned this launch into a mix of hype, memes, and suspicion. One camp was genuinely thrilled, arguing that speed is weirdly underrated and that faster responses make AI tools feel dramatically more useful in everyday work. Another camp zoomed straight past the benchmark chest-thumping and asked the messier question: how much is this going to cost? As one commenter put it, the total lack of pricing gives off strong “if you have to ask...” energy.
And then came the comedy. Cerebras’ giant chips got dubbed “dinner plate chips,” which sounds less like cutting-edge computing and more like something served at a sci-fi buffet. Another commenter immediately started sequel-baiting with “GPT 5.6 Luna Ultrafast when?” Meanwhile, one hot take declared that Google’s speed king, Gemini Flash, may have just been knocked off its throne. In other words: the product launch was fast, but the community reaction was faster.
Key Points
- •Cerebras and OpenAI previewed Ultrafast Mode, a new service tier launching first in the OpenAI API with initial access for a select group of customers.
- •The article says GPT-5.6 Sol on Ultrafast Mode delivers up to 750 output tokens per second without quality compromise.
- •Cerebras cites comparisons showing GPT-5.6 Sol on Ultrafast runs 11x faster than Fable 5 and 5x faster than Opus 4.8 on Fast mode.
- •In Cerebras’ Humanity's Last Exam evaluation, GPT-5.6 Sol Ultrafast completed 2,500 questions in 11 hours and 11 minutes, versus 78 hours and 27 minutes for Claude Fable 5.
- •The article says Ultrafast achieved a 5.6x speedup on GDP-Val and is intended for high-stakes uses such as outage response, cybersecurity, and real-time agent workflows.