What happened
On 9 June 2026, Anthropic released Claude Fable 5, which it calls its most capable model made safe for general use. Anthropic says Fable 5 is state-of-the-art on nearly all of the benchmarks it tested, software engineering, knowledge work, vision, and scientific research, and that its lead over other models grows the longer and more complex a task gets.
The model launched at $10 per million input tokens and $50 per million output tokens, which Anthropic notes is less than half the price of its earlier Mythos Preview. It is available through the Claude API right away, rolling out to subscription plans between 9 and 23 June, and works with Claude Code through the same API.
Fable 5 ships with safety classifiers in a few sensitive areas: when one triggers, the request is answered by Anthropic's next-most-capable model, Claude Opus 4.8, instead. Anthropic says this happens in under 5% of sessions. A sibling release, Claude Mythos 5, is the same underlying model with some safeguards lifted, restricted to vetted cybersecurity professionals.
Why it matters
For a beginner, the headline isn't the benchmark chart, it's the combination of *cheaper* and *steadier on long tasks*. Most early disappointment with AI comes from models that look brilliant for one paragraph and then lose the thread across a multi-step job.
A model that holds its quality over longer work, at a lower price, is the precondition for useful AI agents: assistants that don't just answer one question but read a document, write some code, check their own output, and keep context across all of it. That is the difference between a chatbot and something that can actually take a task off your plate.
It also resets the math. Work you priced out a month ago, summarising long reports, drafting code, triaging support tickets, may now be cheap enough to try.
What to do next
- If the terms are unfamiliar, start with the plain-language entries on what an LLM is and what a token is, token counts are how you'll be billed.
- If you already use Claude or a similar assistant, retry one task that failed before because the model "lost track" halfway through. Long-task reliability is exactly what this release claims to improve.
- If you build software, read the official announcement (linked in Sources) for the API details before switching any production workload, pricing and the 5%-of-sessions safeguard behaviour are worth understanding first.
This briefing summarizes a public, dated announcement from Anthropic and links to its primary sources rather than reporting anything new.