Anthropic launches Claude Fable 5 and Mythos 5: 1M-token context, always-on adaptive thinking, at $10/$50 per million tokens

Anthropic on June 9, 2026, launched Claude Fable 5 and Claude Mythos 5, its most capable publicly released models to date. Fable 5 is immediately available to all users across Claude.ai and the API; Mythos 5—the same underlying model with expanded permissions in select areas—is restricted to Project Glasswing participants, Anthropic's program for vetted cybersecurity professionals and critical infrastructure operators.

What's new

Both models carry a 1 million token default context window, 128,000 maximum output tokens, and always-on adaptive thinking—previously an opt-in extended thinking mode that required explicit activation. Pricing is $10 per million input tokens and $50 per million output tokens, which Anthropic describes as "less than half the price of Claude Mythos Preview."

The models ship with the tokenizer first introduced in Claude Opus 4.7. The same text now produces roughly 30% more tokens compared to pre-Opus 4.7 models, which means developers migrating existing workloads need to re-benchmark token costs and adjust prompt budgets accordingly. Anthropic published an updated token counting API to help measure prompts under the new tokenizer.

A new safety layer runs classifiers on requests and during generation. When a classifier declines a request, the Messages API returns stop_reason: "refusal" and no charge is applied. An opt-in fallbacks parameter—currently in beta on the Claude API and Claude Platform on AWS—lets refused requests re-run on a designated fallback model, billed at that model's rates.

Through June 22, Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans. API model IDs are claude-fable-5 and claude-mythos-5.

Highlighted capabilities from the announcement:

• Software engineering: Stripe reported that Fable 5 "compressed months of engineering into days," including a codebase-wide migration of a 50-million-line Ruby codebase completed in a single day that "would otherwise have taken a whole team over two months"
• Vision: "Fable 5 is the new state-of-the-art model for tasks involving vision," including "rebuilding a web app's source code from screenshots alone"
• Long-horizon autonomy: "Fable 5 and Mythos 5 can work autonomously for longer than any previous Claude models"
• Life sciences: Mythos 5 "with protein design and bioinformatics tools but no human assistance, matches or beats skilled human operators"

Safety classifiers trigger, on average, in less than 5% of sessions. When they fire, the default fallback is Claude Opus 4.8.

Context

Fable 5 arrives as part of a structural naming change. The Opus/Sonnet/Haiku tier system that has defined Claude's lineup since launch is being replaced: "Fable" is the broadly accessible frontier tier; "Mythos" is the restricted-access variant for high-trust, high-risk deployments. This is the first full model family to ship under the new naming scheme.

Project Glasswing—referenced in earlier Anthropic safety research—formalizes a split that has been discussed across the frontier lab landscape but rarely operationalized: a version of the same model with carefully managed permissions for operators in sensitive domains. Mythos 5 participants accept additional terms and are subject to enhanced monitoring.

The fallback billing structure directly addresses one of the most frequently cited friction points for developers evaluating Claude for enterprise use: unpredictable costs from safety triggers. The stop_reason: "refusal" exemption, combined with the optional retry mechanism, gives developers clearer cost modeling.

Why it matters

Anthropic's claim that "Fable 5's capabilities exceed those of any model we've ever made generally available" lands in a moment when Claude Opus 4.8 has already been competitive with GPT-5.5 and Gemini 3.5 on complex tasks. The combination of 1M-token context, always-on thinking, and third-party engineering evidence from Stripe puts this at a credible frontier tier.

Pricing at $10/$50 per million tokens—roughly half what Mythos Preview cost—continues the pattern of frontier-model cost compression that has defined the past 18 months of the market. The tokenizer change cuts both ways: 30% more tokens per prompt means real cost increases for unchanged workloads, which partially offsets the lower per-token rate for teams migrating from Mythos Preview.

The always-on adaptive thinking is the most structurally significant change for developers. Previous extended thinking required opt-in and added both complexity and latency. Making it the default changes the baseline expectation for what every Claude response produces, which matters for evaluating the model against competitors where reasoning modes are still optional overlays.

For the field overall, Fable 5 sharpens competitive pressure on the rest of the frontier. GPT-5.5 and Gemini 3.5 are its direct peers. Anthropic is making a public case that this release closes or narrows whatever gaps existed on long-horizon agentic tasks. Independent benchmark results will determine how that claim holds.