Anthropic Announces Claude Fable 5 and Claude Mythos 5

Anthropic on Tuesday announced Claude Fable 5 and Claude Mythos 5, a pair of frontier AI models the company describes as its most capable ever - and the first time a system from its restricted "Mythos" class has been made broadly available to the public. Fable 5 went live the same day across the Claude API, Claude apps, Claude Code, and every major cloud platform.

"Fable's capabilities exceed those of any model we've ever made generally available," the company wrote in its announcement, positioning Fable 5 as "a Mythos-class model made safe for general use" and "the next generation of intelligence for the hardest knowledge work and coding problems." Anthropic says the model is "thorough, proactive, and tests its own work."

A two-model launch with a deliberate split

The release is structured as a deliberate two-tier launch. Claude Fable 5 (claude-fable-5) is the generally available model, shipped with a real-time safeguard stack covering cyber, bio, and reasoning-extraction risks. Claude Mythos 5 (claude-mythos-5) is, by Anthropic's description, the same underlying model with cyber safeguards lifted - and it is restricted to approved partners in Project Glasswing, the company's cyberdefense program for the US government and critical-infrastructure operators.

The split traces back to April 2026, when Anthropic disclosed Claude Mythos Preview, a model it withheld from general release after internal testing showed superhuman cyber-exploit capability. Roughly 50 vetted partners used that predecessor to find more than 10,000 high- and critical-severity zero-day vulnerabilities, including flaws in every major operating system. TechCrunch noted the company released Fable 5 publicly "days after warning AI is getting too dangerous"; Gizmodo's headline called Fable 5 "a safer version of its 'too dangerous' Mythos AI."

Specs and pricing

Fable 5 ships with a 1-million-token context window and up to 128K output tokens per request. API pricing is $10 per million input tokens and $50 per million output tokens - exactly twice the rate of Claude Opus 4.8, the previous flagship. Subscribers on Pro, Max, Team, and seat-based Enterprise plans get the model included at no extra cost through June 22, 2026, after which it will require usage credits billed at API rates.

The benchmark picture

Anthropic published launch evaluations showing Fable 5 setting the state of the art on nearly every tested benchmark, in several cases by margins NBC characterized as more than 10 percentage points over Opus 4.8.

Benchmark	Fable 5	Opus 4.8	GPT-5.5	Gemini 3.1 Pro
SWE-Bench Pro	80.3%	69.2%	58.6%	54.2%
Terminal-Bench 2.1	88.0%	82.7%	83.4%	70.7%
FrontierCode Diamond	29.3%	13.4%	5.7%	-
GDPval-AA (Elo)	1932	1890	1769	1314
Humanity's Last Exam (with tools)	64.5%	57.9%	52.2%	51.4%
OSWorld-Verified	85.0%	83.4%	78.7%	76.2%

The sweep is not total: GPT-5.5 retains its leads on ARC-AGI-2 and GPQA. But the software-engineering gap is the story of the launch - an 11-point jump over Opus 4.8 on SWE-Bench Pro, and more than double its predecessor's score on FrontierCode Diamond, a benchmark of unsolved frontier coding problems.

Customers report dramatic results

Anthropic's announcement leaned heavily on early-access customer accounts. The most striking came from Stripe, which said Fable 5 migrated a 50-million-line Ruby codebase in a single day - work the company had scoped at roughly two months - and "compressed months of engineering into days."

Cursor CEO Michael Truell called it the "state of the art model on CursorBench." GitHub Chief Product Officer Mario Rodriguez described the model as "a real step forward." Cognition CEO Scott Wu said it scored highest on the company's FrontierBench evaluation, and Replit CTO Fabian Hedin said Fable 5 "understands what builders mean, not just what they type" and frequently "one-shots" complete builds.

Beyond coding, Salesforce's Yusuke Kaji noted that "at the highest effort, [Fable 5] reflects on and validates its own work." Hebbia's Izzy Miller called it the "first to break 90% on our core analytics benchmark... a 10-point jump." Nomura described it as the "strongest finance-first model we've tested," and law firm Davis Polk & Wardwell said a blind review by its lawyers found the model's output "materially different." Anaconda's Peter Wang measured it 25-30% faster than Opus 4.8 on his team's workloads, and Amazon's Luke Anderson reported "more capable engineering in fewer turns."

Anthropic also highlighted scientific results: roughly 10x acceleration on protein and drug-design tasks across 9 of 14 targets, molecular-biology hypotheses preferred by expert reviewers about 80% of the time in blind comparisons, and a week-long autonomous genomics project spanning 138 species that produced a custom machine-learning model outperforming a recent Science publication while being 100x smaller.

Where it ships

In a first for an Anthropic frontier model, Fable 5 reached general availability on every major platform simultaneously: the Claude API, the Claude Platform on AWS, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, and GitHub Copilot. Claude Code users can switch with /model fable.

Safety posture

Fable 5 ships under Anthropic's ASL-3 protections with a CB-1 designation, behind real-time classifiers for cyber, bio, and reasoning-extraction risks. When a session is flagged, an opt-in fallback routes it to Opus 4.8 - something Anthropic says happens in under 5% of sessions, while conceding the system is "stricter than would be ideal... This will be frustrating to some users." The company reports more than 1,000 hours of external red-teaming with no universal jailbreaks found, though the UK AI Safety Institute "made progress toward one." A 319-page joint system card covers both models, and Fable 5's classification as a Covered Model carries a 30-day mandatory data-retention requirement.

The launch sets up an unusual dynamic: the most capable AI model available to the public, sold with an explicit admission that its guardrails will sometimes get in the way - and a more dangerous twin kept behind government-vetted doors.