Anthropic announced two new artificial intelligence models on Monday, Claude Fable 5.2 and Claude Mythos 5.2, pitching them as tools built to take on what the company calls "daunting" technical work. The release, reported by The Verge, extends a model family Anthropic first introduced five months ago and underscores how quickly major AI labs are now iterating on their flagship systems.
Anthropic says the new models show gains in coding, reasoning and vision, and generate responses faster than the prior 5.1 versions of Fable and Mythos. The company claims Claude Fable 5.2 can handle difficult software-engineering problems, coordinate teams of AI agents working on the same task, and sustain long-running projects without losing track of earlier steps. During prerelease testing, Anthropic says the models also made progress on open problems in theoretical physics — a claim that, like the rest of the company's performance figures, comes from Anthropic's own evaluation rather than an independent benchmark.
A two-tier release, again
The naming scheme carries over a structure Anthropic introduced in June, when it first launched Claude Fable 5 and Claude Mythos 5. The two models share the same underlying system but differ in how much is held back. Fable is the broadly available version, wrapped in safety classifiers that route especially sensitive requests to a more conservative fallback model. Mythos carries fewer guardrails and is restricted to vetted cybersecurity and biomedical-research partners through a program Anthropic calls Project Glasswing, reflecting concern that a more capable, less constrained model could also be misused for offensive hacking or harmful biological research.
That the company is already on its third point release in the Fable-Mythos line since June — 5.0, 5.1 and now 5.2 — illustrates a release cadence that has become standard across the industry: ship a capable model, then push incremental upgrades every few months rather than waiting for a single, large jump in capability.
A crowded field
Anthropic is not alone in moving fast. Days earlier, Chinese food-delivery giant Meituan released LongCat-2.5, a mixture-of-experts model with roughly 1.6 trillion total parameters built for long-horizon, multimodal agent work, part of a wave of Chinese labs pushing into the same agentic-coding niche that Anthropic is targeting with Fable 5.2. Meta, Google and OpenAI have all shipped their own updated coding- and reasoning-focused models in recent months, each claiming leadership on one benchmark or another.
That competitive pressure cuts both ways. It is pushing labs to ship real improvements faster, but it also means headline claims — like Anthropic's physics-problem anecdote — arrive well ahead of independent verification. Outside researchers typically take weeks to run their own evaluations after a model's release, and some labs' past capability claims have been revised once third parties got access.
For now, Claude Fable 5.2 is rolling out to existing subscribers and API customers, while Mythos 5.2 remains limited to Project Glasswing partners. Anthropic has not said when, or whether, it plans to loosen that restriction. Given how quickly this model family has already moved from 5.0 to 5.2, the more immediate question for customers may simply be how long 5.2 remains the newest version before the next upgrade arrives.
