Bottom line: Fable 5 sets new benchmarks in software engineering and knowledge work through extended autonomous runtimes, while Mythos 5 offers cybersecurity capabilities without security restrictions.
Anthropic has released Claude Fable 5 as the first generally available Mythos-class variant, surpassing all previous benchmark standards. Parallel to this, Claude Mythos 5 is available with reduced safety measures for cyber defenders as part of Project Glasswing.
Claude Fable 5 is a Mythos-class model that Anthropic has released with safety measures for general use. Its capabilities surpass all previously publicly available Anthropic models and achieve state-of-the-art results in software engineering, knowledge work, vision, scientific research, and other domains. The longer and more complex a task, the greater the advantage over competing models.
Due to the risks posed by this performance – particularly the potential for misuse in cybersecurity – Anthropic has equipped the model with protective mechanisms: requests on sensitive topics are instead answered by Claude Opus 4.8. These safeguards are conservatively calibrated and result in blocks less than 5% on average. Anthropic is working to reduce false positives as new model generations mature. In parallel, the company offers Claude Mythos 5 – the same model with lifted safeguards – to a small group of cyber defenders and infrastructure providers. Deployment begins through Project Glasswing in collaboration with the U.S. government as an upgrade to Claude Mythos Preview. Mythos 5 will soon be accessible through an expanded trusted access program.
In benchmarks, Fable 5 demonstrates strength in software engineering: at Stripe, the model completed a codebase migration in Ruby across 50 million lines in a single day – a task that would otherwise have taken a team two months. On Cognition’s FrontierCode evaluation, Fable 5 achieves higher scores than competing models, even at medium effort levels. In the knowledge work domain, it achieves top results on Hebbia’s Finance Benchmark with improvements in document-grounded reasoning, chart and table interpretation. IMC confirmed Fable 5’s peak performance in trading analytics: fact lookup, conceptual reasoning, root-cause analysis, and expected-value analysis. On vision tasks, Fable 5 is current state-of-the-art: the model precisely extracts numbers from scientific graphics and performs complex tasks such as reconstructing web app source code.
Pricing for both models is $10 per million input tokens and $50 per million output tokens – less than half of Claude Mythos Preview. Anthropic positions this launch as a step toward its goal of rapidly and safely delivering advanced AI capabilities to a broad user base. In practical deployments like Project Glasswing, the models have already assisted in securing critical software; in life sciences research, they contribute to new hypotheses and accelerated drug development.
Source: www.anthropic.com · Published June 8, 2026
Lumi AI News — AI-assisted curation in accordance with Article 50 EU AI Act. Paraphrase and classification through Lumi News Pipeline v1.6.5.