Alex Getmanearly AI signals
Alex Getman @alexgetmancom · · Updated

Tools & updates

Day 1 of OpenAI’s sprint: overall positive

Day 1 of OpenAI’s sprint: overall positive.

Going from ~30 TPS to ~50 TPS is already a meaningful improvement and basically puts it at Fable-level speed.

That said, I think the bar is higher now. I’d like to see at least ~70 TPS, around Opus territory, and ideally 100+.

For frontier AI, 100 TPS should be the standard at this point.

Bar chart comparing output speed in tokens per second for various AI models, including DeepSeek V4.1 (213), Muse Spark 7.3 (187), Claude Sonnet 3.5 (94), Grok 4.7 (91), Claude Opus 5.5 (72), Claude Fable 5.1 (53), GPT-6 Astra (52), and GPT-6f Sol (48).
Fable 5OpenAI
Русская версияMarkdown

Related posts