What happened on June 9, 2026
Anthropic released two models simultaneously: Claude Fable 5 and Claude Mythos 5.
They're the same underlying model. The difference is in the safeguards. Fable 5 is the general-use version — available to everyone, with active protections on cybersecurity, biology, and distillation. Mythos 5 is the version with some of these protections removed, reserved for partners of the government program Project Glasswing.
Why two names? Because Anthropic decided that a model with Mythos-level capabilities cannot be released without filters to the general public. The choice is consistent with the RSP policy the company has maintained for years: the more powerful the model, the more granular the controls.
For enterprise users, the model that matters is Fable 5. And it's available today.
The benchmarks: real numbers, not marketing
Fable 5 tops virtually every benchmark tested across nearly every category. This isn't a vague claim — the results are verifiable and confirmed by third parties.
On coding, Cognition awarded Fable 5 the top score on FrontierCode, even at medium effort. Stripe reported migrating a 50-million-line Ruby codebase in a single day — work that would have taken over two months with traditional methods.
On financial reasoning, Hebbia recorded the top score on its Finance Benchmark — a test designed to evaluate senior-level analysis capabilities. Bloomberg commented that Fable 5 operates at the level of a senior research scientist.
On vision, the model sets a new state of the art. It can extract precise numbers from complex scientific figures, reconstruct source code from web app screenshots, and — a detail that raises a smile — complete Pokémon FireRed using only vision, without maps or external tools.
On long context, it maintains focus across millions of tokens in extended sessions. File-based persistent memory improves performance by 3x compared to Claude Opus 4.8.
How much Fable 5 costs and who can use it
Pricing is more aggressive than expected: $10 per million input tokens, $50 per million output tokens. Less than half of Mythos Preview — with superior performance.
For a business processing documents, analyzing a 30-page contract costs approximately $0.15. A complete financial report with multi-document analysis stays under $2. Costs are competitive with any enterprise alternative on the market.
Availability is immediate on API (model ID: claude-fable-5) and consumption-based Enterprise plans. For subscription plans — Pro, Max, Team — Fable 5 is included free through June 22. From June 23, it will require additional usage credits.
For a complete overview of Claude costs for businesses, we have a dedicated guide comparing all available plans.
What changes compared to Opus 4.8
If your business already uses Claude Opus 4.8, the leap is significant on three fronts.
First: complex, long-running tasks. Fable 5 can operate autonomously for longer periods without losing coherence. This changes the feasibility of projects like codebase migrations, document analysis across hundreds of pages, or compliance audits across entire archives.
Second: token efficiency. Fable 5 uses fewer tokens to achieve the same result. In practice, it costs less for equivalent output — an advantage that compounds with the already competitive pricing.
Third: vision. If you work with scanned documents, charts, tables in PDFs, or screenshots, Fable 5 is in a different league. Data extraction from images goes from being an approximate workaround to a reliable tool.
This isn't an incremental upgrade. It's a category shift in what you can delegate to AI with confidence.
Get updates on Claude and AI for business
One email when there's something worth reading. No spam.
Evaluating Claude for your organization? Learn what it costs or which plan fits best
The safety question: what classifiers are
Anthropic introduced a three-tier safety system for Fable 5, based on dedicated classifiers.
The first covers cybersecurity: it blocks requests about vulnerability discovery and exploitation, offensive hacking, and lateral network movement. Over 1,000 hours of external red-teaming found no universal jailbreak. Zero compliance with harmful requests across 30 tested public jailbreak techniques.
The second covers biology and chemistry: a conservative fallback on any related request. The safeguards are intentionally broad — better to block too much than too little, Anthropic reasons.
The third covers distillation: it detects and blocks attempts to extract the model's capabilities for training competing models.
For businesses that need to meet compliance and GDPR requirements, there's another development: data retention for Mythos-class traffic is 30 days, with human access logging and a prohibition on using data for training.
Mythos 5 and Project Glasswing: the model you can't use (yet)
Mythos 5 is identical to Fable 5 under the hood, but with cyber safeguards removed. Currently it's only available to Project Glasswing partners — a collaboration between Anthropic and the U.S. government for national security.
Anthropic has announced it will soon grant access to selected biology researchers, with bio/chemistry safeguards removed but cyber protections maintained. A broader trusted access program is in the planning stages.
For the European and Italian enterprise market, Mythos 5 isn't relevant today. What matters is Fable 5 — which offers the same reasoning capabilities and performance, with the protections any responsible business would want active.
What companies using it are saying
Testimonials from launch partners speak louder than any benchmark.
Stripe: compressed months of engineering into days. The 50-million-line Ruby migration completed in a day is the most cited case, but it's not an anomaly — it's the direct result of Fable 5's ability to maintain coherence across long tasks.
Cursor (the AI editor): state of the art on CursorBench, with a new class of long-horizon problems that were simply unreachable before.
GitHub: a real step forward in autonomy and reliability across the software development cycle.
Bloomberg: reasoning at senior research scientist level, a clear step beyond Opus 4.8.
DuckDB: apps that previously required 100 prompts are now generated in a single interaction.
These aren't courtesy endorsements. These are companies putting their names on measurable results.
What it means for your business
If you're a business evaluating or already using Claude, here's what changes concretely.
If you're still exploring: Fable 5 lowers the bar for what's feasible. Tasks that previously required constant supervision — complex document analysis, contract review, financial reporting — can now be delegated with a level of reliability that didn't exist before. Now is the time for an AI readiness assessment.
If you're already using Claude in production: reassess your most complex workflows. What didn't work with Opus 4.8 likely works with Fable 5. Migrations, multi-document analysis, tasks requiring hours of reasoning — these are now within the realm of possibility.
If you're in private equity or finance: Hebbia's top score on the Finance Benchmark isn't a footnote. Fable 5 reasons across financial documents at a senior level. For financial modelling, due diligence, and M&A analysis, the leap is tangible.
Maverick AI works with businesses on Claude implementation. If you want to understand how to leverage Fable 5 in your processes, let's talk.