Anthropic just shipped two models that raise the ceiling on what you can build, but they come with a safeguard layer that will affect how you design your integrations.
Claude Fable 5 and Claude Mythos 5 are now available. Fable 5 is the generally available version. It is described as state-of-the-art on nearly all tested benchmarks, with standout performance in software engineering, knowledge work, vision, and scientific research. The headline pattern from Anthropic: the longer and more complex the task, the larger Fable 5's lead over their previous models. That matters if you are building agentic pipelines or multi-step workflows.
The pricing is concrete. Both models are $10 per million input tokens and $50 per million output tokens. Anthropic says that is less than half the price of Claude Mythos Preview, the previous top-tier offering.
Here is the part that shapes your integration decisions. Fable 5 ships with conservative safeguards tuned to prevent misuse in areas like cybersecurity. When a query hits those safeguards, the model transparently falls back to Claude Opus 4.8, Anthropic's next-most-capable model. Anthropic reports this fallback triggers in less than 5% of sessions on average. That is a low number, but non-zero. If your product handles security tooling, vulnerability research, or adjacent topics, you should expect some queries to be silently routed to a different model. Build observability around that from day one.
False positives are an acknowledged reality. Anthropic says the safeguards are intentionally conservative and will sometimes catch harmless requests. They have committed to reducing false positives as more capable models arrive in the coming months.
Mythos 5 is a separate tier. It is the same underlying model as Fable 5, but with safeguards lifted in specific areas. Access is currently restricted to a small group of cyberdefenders and infrastructure providers through Project Glasswing, a collaboration with the US government. Anthropic says it holds the strongest cybersecurity capabilities of any model in the world, and plans to expand access through a broader trusted access program.
The practical signal from early deployments: Project Glasswing has used the models to help secure critically important software. On the life sciences side, the models are generating novel hypotheses and accelerating therapeutics development.
What to do today: If you are integrating Fable 5, instrument your API calls to log which model actually responded. The Opus 4.8 fallback is real and will affect output quality and latency for affected queries. Review your use cases against Anthropic's safeguard areas before launch. If your product genuinely requires unrestricted cybersecurity capabilities, get on the waitlist for the Mythos 5 trusted access program now, since that program is the only path to the full model.