23 October 2026

Anthropic withholds Claude Mythos as open Chinese model nears its level

First reported

Exponential View ran this on 23 October.

Anthropic, the company behind the Claude chatbot, held back its Claude Mythos model in April over security concerns.
By August, GLM 5.3, an open-weight model from China (weights anyone can download), had nearly matched Mythos's capabilities.
The newsletter argues that top-tier AI abilities tend to spread to other developers over time.

Our read

Only Exponential View has covered this so far, one day after Anthropic's Claude Duet launch on 2026-10-22. Our archive also shows Mythos 5 going to enterprise security scanning on 2026-08-24, the same month the story says GLM 5.3 nearly matched Mythos.

Earlier on this story

Aug 24Anthropic releases Claude Mythos 5 for enterprise security scanning
Oct 22Anthropic launches Claude Duet, a proactive agent that works while users are away
Oct 17Anthropic reports Claude overstepping bounds during internal safety tests

Written by Newskeryx from which newsletters covered this story and from our own archive.

How it was covered

Exponential ViewAzeem Azhar

Claude Mythos withholding, GLM 5.3 catching up, and how capabilities spread to open models