Claude Fable 5 and Mythos 5: Anthropic's New Top Tier (And the Catch)

What Anthropic Actually Announced

On June 9, 2026, Anthropic announced Claude Fable 5 and Claude Mythos 5, introducing a new "Mythos-class" tier that the company positions above Claude Opus in capability. The two products share the same underlying model. Fable 5 is the generally available version with extra safeguards on dual-use areas like cybersecurity; Mythos 5 is the same model with some of those safeguards lifted, offered only to approved organizations.

It is worth being upfront about the most important detail. On June 12, three days after launch, Anthropic posted an update: "We are suspending access to Claude Fable 5 and Claude Mythos 5." The page apologizes for the disruption and says the company is working to restore access. So as you read the capability claims below, remember that as of this writing the models are not actually available. We don't know the reason for the suspension from the page text, and we won't speculate.

The pricing, when it was live, was listed at $10 per million input tokens and $50 per million output tokens. Anthropic describes that as less than half the price of the earlier Claude Mythos Preview.

The Capability Claims

Anthropic describes Fable 5 as state-of-the-art on nearly all tested benchmarks, with the gap over its other models widening on longer, more complex tasks. A few specific examples from the page:

Software engineering. Stripe reportedly used Fable 5 to perform a codebase-wide migration in a 50-million-line Ruby codebase in a day, work the company says would have taken a team over two months by hand. On Cognition's FrontierCode evaluation, which checks whether models pass hard coding tasks while meeting production-quality standards, Anthropic says Fable 5 scored highest among frontier models even at medium effort, and is more token-efficient than past Claude models.

Knowledge work. The page cites a top score on Hebbia's Finance Benchmark for senior-level reasoning, with gains in document reasoning and chart and table interpretation. IMC is quoted saying the model did well across its trading-analysis evaluations.

Vision and long-context. Anthropic claims Fable 5 is the new state-of-the-art for vision tasks, including rebuilding a web app's source code from screenshots and beating Pokémon FireRed with a vision-only harness where earlier models needed more scaffolding. On long-running tasks, the company says file-based memory boosted its performance on the game Slay the Spire substantially more than it did for Opus 4.8.

These are vendor claims and partner reports, not independent results. The Stripe and FrontierCode figures are interesting precisely because they speak to the kind of long, messy, real-codebase work that automation builders care about. But treat them as the supplier's framing until third parties reproduce them.

Why This Matters for Builders (and Where It Fits)

For anyone building automations, the relevant theme is duration and autonomy. Anthropic's pitch is that these models can work independently for longer stretches than previous Claude versions, which is the bottleneck for agentic workflows that span many steps. If a model can carry context across a multi-hour task and improve its own output using notes, that changes what you can hand off versus what you have to babysit.

The surrounding product surface is the practical part. Alongside the chat interface, Anthropic offers API access, Claude Code for agentic command-line coding, and Claude Cowork, an agentic desktop app aimed at non-developers. There are also beta agents for Chrome, Excel, and PowerPoint. For automation, the API and Claude Code are the entry points: you'd wire model calls into your own pipelines, or trigger them through orchestration tools. If you're connecting Claude to other systems, platforms like Zapier, Make, and n8n are the usual glue, and OpenAI models remain the obvious comparison point for the same jobs.

The safeguard design is a real consideration for production use. Anthropic says Fable 5 routes some queries to Opus 4.8 instead, and that these filters are tuned conservatively, triggering in under 5% of sessions on average and sometimes catching harmless requests. For an interactive chat that's a minor annoyance. For an unattended automation, an unexpected model substitution or a false-positive block on roughly 1 in 20 sessions is something you'd need to design around, with fallbacks and monitoring.

The Mythos 5 side is narrower. It's aimed at cyberdefenders and infrastructure providers, initially through Project Glasswing in collaboration with the US government, and Anthropic claims it has the strongest cybersecurity capabilities of any model. That's not a tier most teams will touch, and access is gated.

The honest summary: the claims are ambitious, the product lineup around the model is genuinely built for delegation and agentic work, and the access is currently suspended. The sensible move is to watch for the restoration update and for independent benchmarks before planning anything around it. In the meantime, if you want help wiring Claude or any frontier model into real automations, you can browse the provider directory to find people who do this for a living.