Trending on X: Anthropic Ships Claude Opus 4.7 and Admits It Trails an Unreleased "Mythos" Model
What Happened
Opus 4.7 launched across Claude, Bedrock, Vertex, and Microsoft Foundry with an +11% gain on SWE-Bench Pro, 3x vision resolution, and a new "xhigh" reasoning mode at the same $5/$25 per million tokens as 4.6. In a rare move, Anthropic benchmarked it side-by-side against an unreleased internal model called Mythos Preview, seeded only to select cybersecurity and enterprise partners.
My Take
Publishing your "better" private model on the same launch page is a strategic signal, not a humble confession. Anthropic is telling regulators it gates capability voluntarily, telling enterprises there is a premium tier to pay for, and telling rivals that the public frontier is no longer where the frontier actually lives. Expect OpenAI and Google to copy this two-tier disclosure playbook by summer, which quietly ends the era of "best model = what you can buy."
Read Original Source