Claude Fable 5 Explained: Anthropic's First Public Mythos-Class Model — Benchmarks, Pricing, and What Changes for Developers

On June 9, 2026, Anthropic released Claude Fable 5, a new model tier that sits above the Opus family and is, by Anthropic's own description, more capable than any model the company has ever made generally available. It launched alongside Claude Mythos 5 — the same underlying model with certain safeguards lifted, restricted to a small group of cyberdefenders and infrastructure providers. This guide covers what Fable 5 is, how it performs, what it costs, what changes in the API, and who should upgrade.

What is Claude Fable 5?

Claude Fable 5 is Anthropic's first publicly available Mythos-class model. Until now, Anthropic's most powerful internal models (the Mythos line) were not offered to the public. Fable 5 is that class of capability, shipped with safeguards that make it safe for general use.

The headline claims from Anthropic's announcement:

• State-of-the-art on nearly all tested benchmarks of AI capability, with exceptional performance in software engineering, knowledge work, vision, and scientific research.
• The longer and more complex the task, the larger Fable 5's lead over other Claude models — it is built for long-horizon, autonomous work.
• A safeguard fallback system: queries on a small set of sensitive topics are answered by Claude Opus 4.8 instead. Anthropic says these safeguards are tuned conservatively and trigger in fewer than 5% of sessions.

Fable 5 vs Mythos 5: why two models?

The launch is deliberately split in two:

• Claude Fable 5 is the public release — full capability with safety guardrails in place, available to everyone through the API and Claude products.
• Claude Mythos 5 is the same underlying model with safeguards lifted in some areas. It is being deployed only to a vetted group of cybersecurity defenders and critical-infrastructure providers through Project Glasswing, a collaboration with the US government.

This is a notable shift in how frontier labs handle dual-use capability: rather than holding the model back entirely or releasing it unrestricted, Anthropic is splitting access by risk profile.

Benchmarks: how much better is Fable 5?

The most-cited number from launch coverage is agentic coding performance on SWE-Bench Pro:

An 11-point jump over Opus 4.8 — itself the strongest coding model of the previous generation — is the largest generation-over-generation gain Anthropic has shipped. Early enterprise signals point the same way: analytics company Hex reported Fable 5 is the first model to break 90% on its internal benchmark of complex, long-running analytical tasks.

As always, treat vendor-reported benchmarks as directional until independent evaluations land. But the consistent theme across launch coverage is that Fable 5's advantage grows with task length and complexity — short prompts won't show the gap; multi-hour agentic runs will.

Pricing: $10 input / $50 output per million tokens

Fable 5 costs exactly double Opus 4.8, and Anthropic notes it is less than half the price of the earlier Claude Mythos Preview:

Prompt caching matters more than ever at this price point: cached input is discounted roughly 90%, which is the difference between Fable 5 being a premium tool and an impractical one for long agentic sessions that re-read large contexts on every turn.

What changes in the API

Fable 5 (model ID claude-fable-5) keeps the same API surface as Opus 4.7 and 4.8, with the now-familiar modern constraints — plus one new one:

• Adaptive thinking only. The old thinking: {type: "enabled", budget_tokens: N} form returns a 400 error. Use thinking: {type: "adaptive"} and let the model decide when and how deeply to reason.
• New in Fable 5: an explicit thinking: {type: "disabled"} also returns a 400. If you want thinking off, omit the thinking parameter entirely.
• No sampling parameters. temperature, top_p, and top_k are removed and will 400. Steer behaviour through prompting instead.
• Effort levels. The output_config.effort parameter (low through xhigh and max) is the main lever for trading cost against depth. For coding and agentic work, high or xhigh are the recommended starting points.
• 128K max output tokens and a 1M+ context window, with server-side compaction available for conversations that outgrow it.

If your code already runs on Opus 4.7 or 4.8, migrating is essentially a model-ID swap — just make sure no code path sends an explicit thinking: {type: "disabled"}.

Where you can use it today

Fable 5 launched with unusually broad day-one availability:

• Claude API (claude-fable-5) and the Claude apps
• Claude Code, where it is selectable as a model for agentic coding sessions
• GitHub Copilot, generally available from launch day
• AWS, Google Cloud, and Microsoft Foundry for cloud-platform customers
• Enterprise plans on consumption-based pricing

Should you upgrade?

A practical way to decide:

• Upgrade now if you run long-horizon agentic workloads — multi-hour coding sessions, autonomous research, complex migrations. That is exactly where Fable 5's lead is largest, and where higher per-token cost is often offset by fewer wasted turns.
• Test first if you run shorter interactive workloads. For chat, extraction, and classification, Opus 4.8 at half the price — or Sonnet 4.6 at less than a third — may remain the better value.
• Stay put if your pipelines depend on temperature or fixed thinking budgets and you haven't yet done the adaptive-thinking migration. Do that migration on Opus 4.8 first, then the Fable 5 swap is trivial.

FAQ

Is Claude Fable 5 a replacement for Opus?

No — it's a new tier above Opus. Opus 4.8, Sonnet 4.6, and Haiku 4.5 all remain available, and Opus 4.8 actually serves as the fallback model inside Fable 5's safeguard system.

What does "Mythos-class" mean?

Mythos is Anthropic's internal line of most-capable models that were previously not released publicly. Fable 5 is the first model of that class made generally available, with safeguards; Mythos 5 is the restricted-access variant with some safeguards lifted.

What is Project Glasswing?

A collaboration between Anthropic and the US government to deploy Claude Mythos 5 to a vetted group of cyberdefenders and critical-infrastructure providers — giving defensive teams access to frontier capability that isn't broadly released.

How often do the safeguards kick in?

Anthropic says the fallback to Opus 4.8 triggers in fewer than 5% of sessions, and that the safeguards are deliberately tuned to be conservative at launch.

The bottom line

Claude Fable 5 is the most significant Anthropic release since the original Opus: a genuinely new capability tier, priced at a premium but made practical by prompt caching, with the biggest gains landing exactly where the industry is heading — long, autonomous, agentic work. The split release with Mythos 5 is also a template other labs will be watching: full capability for vetted defenders, safeguarded capability for everyone else.