Will an AI model use more than 1e28 FLOPS in training before 2026? | Manifold

Will an AI model use more than 1e28 FLOPS in training before 2026?

5

1kṀ785

Dec 31

24%

chance

1D

1W

1M

ALL

Resolution source: Epoch AI's list of notable AI models. I will check this source on January 1st, 2026, to see whether there is a model that uses more than 1e28 FLOPS https://epoch.ai/data/notable-ai-models

AI models do not only include LLMs, but also other types of AI models that are mentioned in the resolution source

As of market creation, the biggest LLM model is Grok 3, with 4.6e26 FLOPs of training compute

Technical AI Timelines

Get

1,000

to start trading!

Sort by:

Don't see how this is possible. Stargate won't be fully underway until Mid 2026, and if they manage a partial training run with, say 200k GB200s that runs twice as long as Grok 3, that's looking at ~3.68e27 FLOPS. I think it's a similar-ish story for Project Rainier and even for Google.

Related questions

At least one of the most powerful neural nets at end of 2030 will be trained using 10^26 FLOPs

+8% 1d96% chance

Will the largest machine learning training run (in FLOP) as of the end of 2025 be in the United States?

Will an AI achieve >85% performance on the FrontierMath benchmark before 2028?

Will a machine learning training run exceed 10^27 FLOP in China before 2028?

Will a machine learning training run exceed 10^26 FLOP in China before 2028?

At least one of the most powerful neural nets at end of 2026 will be trained using 10^26 FLOPs

+10% 1d96% chance

At least one of the most powerful neural nets at end of 2026 will be trained using 10^27 FLOPs

Will a machine learning training run exceed 10^26 FLOP in China before 2026?

Will there be an announcement of a model with a training compute of over 1e30 FLOPs by the end of 2025?

How much FLOP will be used to train the best language model with freely available weights on July 1, 2025?

Related questions

At least one of the most powerful neural nets at end of 2030 will be trained using 10^26 FLOPs

At least one of the most powerful neural nets at end of 2026 will be trained using 10^26 FLOPs

Will the largest machine learning training run (in FLOP) as of the end of 2025 be in the United States?

At least one of the most powerful neural nets at end of 2026 will be trained using 10^27 FLOPs

Will an AI achieve >85% performance on the FrontierMath benchmark before 2028?

Will a machine learning training run exceed 10^26 FLOP in China before 2026?

Will a machine learning training run exceed 10^27 FLOP in China before 2028?

Will there be an announcement of a model with a training compute of over 1e30 FLOPs by the end of 2025?

Will a machine learning training run exceed 10^26 FLOP in China before 2028?

How much FLOP will be used to train the best language model with freely available weights on July 1, 2025?

© Manifold Markets, Inc.•Terms + Mana-only Terms•Privacy•Rules