Skip to contentNewskeryx

3 November 2026

DeepSeek's new model matches GPT 5.5 on a major AI benchmark

First reported

Latent Space ran this on .

  • DeepSeek's V4.1F-1017 model scored the same as OpenAI's GPT 5.5 on the Epoch Capabilities Index, a broad test of AI ability.
  • Anthropic's Mythos Preview has not been matched by any Chinese model for roughly seven months, according to the newsletter's commentary.
  • The score is contested: critics question how the index weights its tests.

How it was covered

Latent Spaceswyx & Alessio

Epoch index scores, gap to Mythos Preview, weighting critique, benchmark spikiness, reward-hacking concerns