Anthropic just shipped two versions of its newest model, and only one of them is actually available to you. Claude Fable 5.2 is live today across Anthropic's platforms and cloud partners. Claude Mythos 5.2 exists too, but you'll only get your hands on it if your organization goes through Anthropic's verification programs [1]. Both come with a system card spelling out safeguards and known limitations [2], and the release has already picked up a label from outside observers: “jagged AGI,” a phrase Ethan Mollick coined to describe capability that's uneven rather than uniformly superhuman [3].
Here's the part that actually matters for how you work this week. Fable 5.2 is built for ambitious, open-ended technical work rather than everyday drafting or quick lookups [1]. Find one real but bounded project, point Fable at it, and keep a human reviewing the output closely from day one.
Fable 5.2 and Mythos 5.2: Same Model, Different Doors
Fable and Mythos are two configurations of the same underlying model [1]. Fable is the one you'll actually touch: it's available now, carries the API model ID claude-fable-5-2, and ships with extra safeguards around cybersecurity and biology topics [1]. Mythos has more permissive access, but that access is gated behind Anthropic's verification programs, so it's reserved for organizations that go through that process [1]. Fable is the public version with guardrails tuned for broad use. Mythos is the more open variant kept for vetted research and enterprise contexts [1].
Pricing on Fable runs $8 per million input tokens and $40 per million output tokens at standard speed [1]. There's also a faster mode at $16 input and $80 output per million tokens, which Anthropic says runs up to 2.5 times quicker than standard [1]. That speed premium is worth knowing about if you're running Fable inside any pipeline where a person is sitting there waiting on the result.
Where Fable Fits Next to Opus
Anthropic presents Fable as a complement to its flagship [1]. The company still wants Opus 5.5 as your daily driver, and positions Fable as the model you reach for on the harder, messier problems, or deploy as an Advisor or Orchestrator inside multi-model workflows [1]. Anthropic's own benchmarks show Fable ahead of Opus 5.5 in most categories, but that's the company grading its own test, so treat it as a vendor claim rather than independent verification [1]. The practical split: Opus for routine work, Fable for the edge cases where you need more reasoning headroom and you're willing to review the output carefully.
Jagged AGI: A Useful Frame, Not a Verdict
Ethan Mollick reviewed the release independently and called it “almost” AGI, or “jagged AGI” [3]. His examples span hardware hacking, sketching out a business concept, humanities teaching, academic research support, and building large creative 3D-world projects [3]. That's a genuinely wide range, and it's worth taking seriously.
But Mollick is just as specific about the gaps [3]. He reports the model can still miss calls on taste, repeat itself, run on well past the point of usefulness, or spin up its own self-created bureaucracy, layers of process nobody asked for that a person has to notice and cut [3]. His takeaway: you're still the one setting direction and deciding when the output is actually good enough [3].
Anthropic's system card backs this up [2]. The company says plainly that Mythos doesn't come close to substituting for its research staff, and that it doesn't believe the model crosses its threshold for automated AI R&D acceleration, while admitting real uncertainty about that judgment [2]. This is a capability jump worth testing. Mollick's “jagged AGI” label describes its uneven strength and the human judgment it still needs [3].
What to do this week
- Pick one scoped, genuinely hard problem you've been putting off, something open-ended and technical rather than routine, and run it through Fable as a bounded experiment with a clear start and stop point.
- Spot-check a sample of the outputs before anything leaves draft status, and watch specifically for the failure modes Mollick flags: repetition, excessive length, and scope drift past what the task actually needed.
- Run the same task through Opus 5.5 for comparison, so you've got a real basis for deciding which model earns a permanent spot in your workflow and which stays reserved for the hard cases.
If this launch has you thinking about how to get more from high-capability models, the Prompt Engineering for ChatGPT course is a strong next step. It teaches you how prompt structure and context shape what a model produces, so you can direct these systems with more confidence.*
Sources
- Anthropic, Introducing Claude Fable 5.2 and Claude Mythos 5.2
- Anthropic, System Card: Claude Fable 5.2 & Claude Mythos 5.2
- Ethan Mollick, Fable 5.2 and Jagged AGI