Meta's Muse Wonder Drops as Character.AI Files for Bankruptcy

Meta has released its first-ever frontier AI model, Muse Wonder, directly competing with OpenAI and Anthropic for the first time. The host of this episode calls it "really exciting" because Meta had previously lagged behind with open-source offerings that weren't worth actively using. Now he says he's "actually actively using and recommending Meta for certain circumstances." On the personal superintelligence index, Muse Wonder Max scored 59, beating every competitor. On the Harvey Lab benchmark it hit 13, again topping the field. More practically, on the DeepSWE 1.1 benchmark—which measures how well an AI can understand and work with code repositories—it scored 79%.

The model's strength comes from Meta's specialty: deep understanding of user data. As the host puts it, "they got a file on every person" and are generating as much data as possible, which makes the AI incredibly useful for understanding context. The free MetaMuse tool can now run through emails, sort them, delete spam, forward messages to team members, and draft replies. The host uses it every morning and says, "for the first time ever, it made it feel doable." The top-tier version (Muse Wonder Max) is only available to paying subscribers of Muse Code, their coding platform.

Ukrainian Drone Strikes Knock Out Russian AI Training