Anthropic's Claude 5, codenamed Fable, is the subject of this episode, with hosts Josh Kale (an Anthropic contractor) and Ejaaz breaking down benchmark results, capability demos, and the specific restrictions Anthropic has hardcoded around biology, chemistry, and cybersecurity.
The episode is worth reading in full because the demos go beyond standard text tasks: visual reasoning, live gaming, and enterprise workflow automation are shown in practice, not in theory. The benchmark section at 13:58 puts numbers to the claims, and the 'Mythos Model' segment starting at 25:37 addresses something distinct enough to get its own name and 12 minutes of runtime.
Pricing and compute limits land at 25:37, followed by a segment on long-horizon workflows, which is where the real enterprise argument lives. The tension throughout is the same one Anthropic has always carried: stronger capabilities paired with harder restrictions, and whether those two things can scale together.
[WATCH ON YOUTUBE →]