Original caption
Anthropic's unreleased Model 4 just showed up in a METR report, and staff say it makes them about 12x more productive. METR is an independent group that tests AI models. Model 4 is the codename for an internal model that finished training around October, so none of us can use it yet. I love building with Claude, and that's the part that's crazy to me. Not going to lie, I wouldn't trust the 12x. It's staff rating themselves in Slack polls, and METR says it's weak evidence. METR's own estimate is that Anthropic is improving its models about 2x faster than before, and it guesses around 80% of that speedup comes from AI. It could be anywhere from 1.3x to 3x, and Anthropic redacted a lot of the evidence. Do you believe the 12x? Anthropic Model 4, METR report, Anthropic internal models, unreleased Claude model, Claude, Claude Mythos, Claude Opus, AI productivity, AI uplift, AI building AI, AI research, Anthropic news, AI news