Claude Fable 5 Launch: Mythos-Class AI Goes Public, Free Until June 22
Claude Fable 5 is the Mythos-class AI model that Anthropic released to the public on June 9, 2026. Mythos-class is a new tier above Opus, the previous flagship line, until now restricted to select cyber-defense organizations working with the US government. If you pay for Claude, this is about your wallet too: Fable 5 is included free only through June 22.
What Makes Claude Fable 5 Different
In short: institution-grade power, with guardrails. When a request touches risky ground like hacking or biology, a classifier (a separate AI that screens dangerous queries) hands the answer to lower-tier Opus 4.8 instead; over 95% of sessions never trigger it. API pricing is $10 per million input tokens and $50 for output (tokens are the text units AI processes) - double Opus 4.8. Stripe reports it finished a 50-million-line codebase migration in one day - two months of team work, by Stripe's estimate.
Benchmark Results: 21.7 Points Ahead of GPT-5.5
On SWE-bench Pro, which scores models on fixing real GitHub code issues, Claude Fable 5 hit 80.3%. Opus 4.8 scored 69.2%, OpenAI's GPT-5.5 58.6%, and Google's Gemini 3.5 Flash 55.1% (per BenchLM). The lead holds beyond coding: 1,932 on the GDPval knowledge-work test versus 1,769 for GPT-5.5, and 85.0% on OSWorld, the computer-use test.
The June 23 Switch to Usage Credits
Anthropic's official schedule is explicit. Through June 22, Claude Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans. On June 23 it leaves those plans, and continued use requires usage credits (a prepaid usage balance). The company calls it a temporary capacity measure: the free window may be extended, the goal is to restore Fable 5 as a standard plan feature, and changes will be announced ahead of time.
How to Play the Free Window
The rational move: treat these two weeks as a free trial. Run Claude Fable 5 on your actual workload - coding, reports, analysis - and measure the difference before June 22. Then decide whether credits are worth buying. Analysts expect this pattern, top-end AI carrying a surcharge on top of subscriptions, to spread as compute costs soar - a preview of how your AI bill may be structured next.
Key Takeaways
① Mythos goes public - the institutions-only tier opened to everyone, with guardrails, on June 9.
② Benchmark sweep - 80.3% on SWE-bench Pro, ahead of GPT-5.5 (58.6%) and Gemini 3.5 Flash (55.1%).
③ Free until June 22 - from June 23 paid plans need usage credits; temporary, with advance notice promised.
The most capable AI is now open to everyone - and so is a new question: what is it worth to you? The next two weeks are the cheapest way to find out.
👉 Anthropic Files for IPO Ahead of OpenAI - $965B Valuation, 2026 - also worth a read.
📌 Sources: Anthropic, TechCrunch, EdenAI, BenchLM (2026)
댓글
댓글 쓰기