Laurenceo lowers 2027 frontier compute forecast to 22%

Forecast MoveLaurenceo Intelligence21 November 2026

The largest verified training run found so far is about 9.4×10²⁶ FLOP, which would need to rise roughly 2.7 times before the deadline 1.

51.5%22%Lowered on 21 November 2026 · resolves by 30 April 2027

Laurenceo has lowered its forecast that Epoch AI will list a frontier model trained with 2.5×10²⁷ FLOP or more by 30 April 2027. The probability fell from 51.5% to 22% on 21 November 2026. Resolution requires a verified value published by Epoch AI or another primary source, dated on or before that deadline.

Why It Moved

The largest verified run found so far is GPT-6 Astra at about 9.4×10²⁶ FLOP, below the threshold 1. The latest recorded step-up was about 1.9 times the previous record, so reaching the threshold would need a larger or repeated jump 3. Many new models are released without any compute estimate, so a qualifying run may never be posted 2. Trend lines fitted to a few points can also overshoot, and the opening 52% fit rested on an out-of-date latest value. The GPT-6 Astra figure comes from a secondary site and has not been verified.

Laurenceo would raise the forecast on an Epoch AI compute estimate above 1×10²⁷ FLOP for a new frontier model. Early 2027 launches from OpenAI, xAI, Google or Anthropic that publish large training compute would also be watched closely. A further record near 1×10²⁷ FLOP would push the probability lower.

Evidence

The Forecast

Will Epoch AI list a frontier model trained with 2.5 x 10^27 FLOP or more by April 30, 2027? The full reasoning, every revision and the resolution rule are on the forecast page. Laurenceo publishes a story whenever one of its forecasts moves 10 points or more in a day.

More Analysis

Laurenceo Intelligence is the daily AI intelligence briefing published in Bermuda by Laurenceo and edited by Jorai Laurenceo. Every story is researched, source-verified and published each morning; how it works. General information only, not legal, regulatory or financial advice.