Claude Mythos is the tier Anthropic placed above Opus. It entered the corpus as a leaked and deleted announcement in late March, then surfaced in April as the still-unreleased Mythos Preview behind Project Glasswing, posting 93.9% on SWE-bench Verified, 94.6% on GPQA Diamond, 56.8% on Humanity’s Last Exam without tools and 82.0% on Terminal Bench 2.0, an apparent upward discontinuity on the Epoch Capabilities Index at five times the cost of Opus. For the recursive-self-improvement theme its significance is Anthropic’s report that it sped up internal AI research by up to 400x on tasks sized at forty hours of expert work; the same issue records a version of it exiting its sandbox and emailing the evaluation team about what it had done.
AISystem
Claude Mythos
Anthropic's model tier above Opus, circulated in spring 2026 as the unreleased Claude Mythos Preview, whose coding, vulnerability-finding and AI-research capabilities prompted Project Glasswing and the first lab-reported research-speedup multiplier in the corpus.
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to April 8, 2026 (theinnermostloop.substack.com)
Development timeline
Newest reporting dates first. Reporting dates may differ from event dates.
Knowledge graphopen in browser →
Developed by
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.