On June 9th, 2026, Anthropic released Claude Fable 5, the first publicly available model from its Mythos tier. This has been a long awaiting model around safety. And just recently, the May 2026 Ramp AI Index, drawn from corporate card and invoice data across more than 50,000 US companies, shows Anthropic at 34.4% of business adoption, ahead of OpenAI at 32.3%.
Here are six capabilities that matter.
1. Mythos Fable 5 Performance that rewrites what’s possible in code
Fable 5 scored 80.3% on SWE-bench Pro, a benchmark of real software-engineering work. Opus 4.8 reached 69.2%. GPT-4 sits at 58.6%. The gap widens as problems get longer.
Stripe used Fable 5 to migrate a 50-million-line codebase in a day which that work would have taken a team two months. The model doesn’t break tasks into model-sized chunks anymore. It holds the entire project in mind, plans autonomously, runs for hours or days, and validates its own work.
2. Mythos Extended reasoning without cognitive collapse
Where previous models hit a wall on long chains, Fable 5 sustains reasoning across sprawling problems.
Some examples:
- Legal work climbs to 13.3% versus GPT-4’s 2.1%.
- Multidisciplinary reasoning hits 64.5% with tools.
- Biology reaches 83.9% on human-solved tasks.
This matters for anyone running knowledge-heavy workloads: contract analysis, financial audits, scientific research, medical reasoning. The model doesn’t hallucinate under load; it compounds precision across thousands of tokens.
3. Agentic autonomy that actually closes loops
Fable 5 powers agents that plan, execute, and validate without human intervention between steps.
Box, the cloud storage company, says agents will use it immediately for financial documents, contracts, and life-sciences research. The model understands context deeply enough to catch its own errors, backtrack when needed, and explain why it changed course.
For teams deploying autonomous workflows, this is the difference between a pilot and production infrastructure.
4. Mythos Has Vision and knowledge work integrated
The model combines state-of-the-art vision with knowledge reasoning.
Knowledge work (GDPval-AA) scores 1932 versus Opus’s 1890. Vision (GDPdf) reaches 29.8%.
This combination matters for document processing, scientific-paper analysis, and any task that requires reading visual information and extracting semantic meaning simultaneously.
The model doesn’t misread charts or miss details buried in dense layouts.
5. With Mythos, Science acceleration Compounds Research Velocity
The unrestricted model is only available to vetted partners, and those partners are already using it on bleeding-edge science. In drug design, the model accelerated parts of the pipeline roughly tenfold.
For genomics, Mythos 5 trained a custom model that beat a recently published one despite being 100 times smaller. The model generates novel molecular-biology hypotheses that scientists prefer about 80% of the time over Opus-class output. This is research acceleration at scale.
6. Mythos Enterprise pricing that doesn’t require a second mortgage
Both Fable 5 and Mythos 5 cost $10 per million input tokens and $50 per million output tokens. That’s less than half Mythos Preview’s pricing. With companies concerned about the AI Cost Crisis, this is welcomed news.
The model is live on the Claude API and consumption-based enterprise plans now. On subscription plans, according to Anthropic, (Pro, Max, Team, seat-based Enterprise), it’s free through June 22; after that, usage requires credits.
Who’s Already Building on Mythos
Box announced immediate integration of Fable 5 into Box Agents for financial documents, contracts, and life-sciences research. Aaron Levie, the founder and CEO, announced this move on Linkedin. Stripe deployed it for codebase migrations. Lovable’s CTO said apps that once required a hundred prompts can now launch in one shot. A cohort of biology researchers is joining to run Mythos 5 on discovery work.
The architecture matters: Fable 5 and Mythos 5 are identical under the hood.
What separates them is safety filtering. Fable 5 deploys classifiers that detect requests touching cybersecurity, biology, chemistry, or model-extraction attempts and hands them to Opus 4.8 instead. Mythos 5 removes those guardrails for trusted partners. The filters trigger in fewer than 5% of sessions, so they catch edge cases without breaking ordinary work.
All Mythos-class traffic carries a 30-day data-retention policy for safety monitoring; Anthropic commits the data won’t be used for training.
Anthropic opened its most powerful tier to the public.
The move suggests the company’s confidence that Mythos is safe enough for production and that the market is ready for reasoning at this depth.
