Tech Reporter Scoop: Anthropic to Release Public Version of 'Mythos' Model Tomorrow
Claire Weston
Tech reporter Alex Heath says Anthropic plans to release Fable 5 tomorrow — a public version of its internal Mythos model, built for sustained multi-step tasks. This means the AI race is shifting from 'smarter answers' to 'longer, more reliable work.'
What is Fable 5, and how does it relate to Mythos?
Fable 5 is the public release of Mythos, with minor adjustments before general availability.
Mythos first surfaced on March 26 after a CMS configuration leak reported by *Fortune*; its internal codename was Capybara. Anthropic confirmed it was testing a model representing a "step-change in capability" — calling it "the strongest model we have ever built."
In plain terms = Mythos has been running with a closed group of partners for over two months. Now Anthropic is repackaging it as Fable 5 for everyone.
What's the core upgrade — and why do 'long-horizon, multi-turn tasks' matter?
Fable 5's headline improvement is in long-horizon, multi-turn tasks — the ability to work through complex, extended chains of operations without losing coherence.
This means → the competitive frontier is moving from "how smart is a single answer" to "can the model reliably work for hours or days across a sequence of related tasks."
This reflects a growing industry consensus: the real bottleneck for AI replacing human labor is not comprehension — it is sustained, stable execution over time.
How does the public version differ from the partner version?
The public Fable 5 will ship with stricter safety guardrails, particularly around network operations, compared to the partner version.
The partner version is available through Project Glasswing — Anthropic's cross-industry cybersecurity collaboration framework — with roughly 50 initial partners including AWS, Apple, Microsoft, Google, NVIDIA, and CrowdStrike.
In plain terms = same underlying model, but partners get broader permissions and fewer restrictions. The public version matches on capability but runs inside a tighter safety fence.
What do the Mythos performance numbers tell us?
Mythos Preview pricing: $25 per million input tokens, $125 per million output tokens — the most expensive tier among publicly available models.
Benchmark scores: 97.6% on USAMO 2026, 93.9% on SWE-bench Verified.
In security applications, Mythos has uncovered more than 10,000 high-severity or critical zero-day vulnerabilities — software flaws exploitable by attackers before any patch exists — including some that had gone undetected for up to 27 years.
This means → the model has proven real-world value well beyond exam benchmarks, particularly in cybersecurity.
What's the timeline — and what comes next?
March 26: Mythos exposed via configuration leak; Anthropic confirms testing.
April 7: Mythos Preview officially launched, restricted to Project Glasswing partners.
May 28–29: Anthropic ships Claude Opus 4.8 and signals it will push Mythos-class models to all customers "within weeks."
Tomorrow (per report): Fable 5 goes public. This reflects an accelerating cadence — leak to general release in under three months.
Content is for reference only, not financial advice.