Skip to contentNewskeryx

3 November 2026

DeepSeek's V4.1F-1017 scores level with GPT 5.5 on one index

First reported

Latent Space ran this on .

  • DeepSeek, a Chinese AI company, has a model called V4.1F-1017 that scored 159 on the Epoch Capabilities Index.
  • That score matches GPT 5.5, a model from OpenAI, on the same index.
  • The Epoch index is a single combined score across many tests, so one number can hide uneven performance.

How it was covered

Latent Spaceswyx & Alessio

Epoch index scores, gap to Mythos, weighting critique, spikiness on evals, reward-hacking behavior