Google, Marvell reportedly developing AI chips together
The AI race is moving quickly.
And as quickly as companies develop new models, raise new funding, and make new hires, so too does the tech industry’s view of the field’s key players shift.
So, after weeks as the industry’s darling, Anthropic has seen this past week defined by its perceived shortcomings.
On Thursday, Anthropic released its latest model, Claude Opus 4.7.
Notably, this is not its most advanced model — that is Mythos Preview, which it has released to a select number of partners as part of Project Glasswing.
The challenge, however, is that users have noticed.
Tae Kim, author of the newsletter Key Context and a recent book on Nvidia, compiled a series of complaints from notable commentators on X, highlighting a variety of perceived deficiencies in Opus 4.7.
Gergely Orosz, who writes the Pragmatic Engineer newsletter, also called Opus 4.7 “surprisingly combative.”
Orosz has also been among those in the tech space, noting that Claude has notably struggled to keep up with tasks in recent weeks.
Anthropic’s compute disadvantage — that is, the amount of raw computing power available to the company — relative to OpenAI has become a major point of discussion as adoption of Claude, specifically Claude Code, has ramped in recent weeks.
This also comes as the company’s rollout of Mythos Preview continues to confound some commentators, with a description of its capabilities — but a less transparent view of its benchmarks — making the cybersecurity vulnerabilities reportedly surfaced by Mythos feel like a blend of PR spin and a true AGI moment.
