Estimate: V4.1 Pretraining Took ~5e24 FLOPs

Citing an estimate from Astra, V4.1 models reportedly required 5e24 FLOPs for pretraining, slightly above V3, equivalent to 3.5 million H100 hours or two weeks on 4,000 B300 GPUs.

2026-09-10 ~ 2026-09-10 · 2 related posts

• Gemini V4.1 pretraining estimated at ~5e24 FLOPs, two weeks on 4K B300s — teortaxesTex · 2026-09-10
• Back-of-Envelope Estimate Puts DeepSeek V4.1 Pretraining at ~5e24 FLOPs — teortaxesTex · 2026-09-10