Could you also include Opus 4.7 and its place on the AECI trend into your analysis? Opus 4.7 CONTINUED the linear trend, meaning that Mythos likely didn’t accelerate the R&D enough to change Opus’ capabilities.
Anthropic’s statement on Opus’ abilities
Claude Opus 4.7 lands on the pre-Mythos Preview trend. Fitting the linear trend on the Anthropic frontier through Claude Opus 4.6 (slope ≈ 13.6 AECI/yr, n=8), Claude Opus 4.7 sits approximately +1.0 AECI above that line—within error bars of the historical trend. By contrast, Claude Mythos Preview sits approximately +5.8 AECI above the same line. Claude Opus 4.7 does not advance the capability frontier (Claude Mythos Preview, released earlier, scores higher), so it does not affect the slope-ratio calculation directly.
Thanks, that’s a good point. I wrote most of this article before Opus 4.7 was released, and haven’t integrated all new info provided. I’ll add some discussion on this when I get back to my computer.
Could you also include Opus 4.7 and its place on the AECI trend into your analysis? Opus 4.7 CONTINUED the linear trend, meaning that Mythos likely didn’t accelerate the R&D enough to change Opus’ capabilities.
Anthropic’s statement on Opus’ abilities
Claude Opus 4.7 lands on the pre-Mythos Preview trend. Fitting the linear trend on the Anthropic frontier through Claude Opus 4.6 (slope ≈ 13.6 AECI/yr, n=8), Claude Opus 4.7 sits approximately +1.0 AECI above that line—within error bars of the historical trend. By contrast, Claude Mythos Preview sits approximately +5.8 AECI above the same line. Claude Opus 4.7 does not advance the capability frontier (Claude Mythos Preview, released earlier, scores higher), so it does not affect the slope-ratio calculation directly.
Thanks, that’s a good point. I wrote most of this article before Opus 4.7 was released, and haven’t integrated all new info provided. I’ll add some discussion on this when I get back to my computer.