Anthropic's Claude Fable 5 is now available to the public — the first Mythos-class AI model that anyone can access, and the most capable model Anthropic has ever released for general use. It dropped on June 9, 2026, and it changes the baseline for what AI can do across software engineering, knowledge work, scientific research, and security.
At Safeguard, we have already integrated Claude Fable 5 into our Multi-Agent TAOR Deep Think AI Engine. Zero-day discovery, vulnerability remediation, and supply chain security powered by Fable 5 are live for all customers today.
Here is the full picture — what Fable 5 is, what its numbers mean, and why it matters for every security program running in 2026.
TL;DR: Claude Fable 5 is Anthropic's first publicly available Mythos-class model. It scores 80.3% on SWE-Bench Pro, 88% on Terminal-Bench 2.1, and 59% on Humanity's Last Exam. Available on Claude API, Amazon Bedrock, and GitHub Copilot. Safeguard has already integrated Fable 5 natively for continuous zero-day discovery and agentic remediation.
What Is Claude Fable 5?
Claude Fable 5 is Anthropic's most capable publicly available AI model, released on June 9, 2026. It is the first model from Anthropic's Mythos tier — the capability class above Opus and Sonnet — to be made generally available to the public.
Where the full Mythos model (Claude Mythos Preview) remained restricted to Project Glasswing partners for two months due to its raw offensive security capabilities, Fable 5 is the version of that capability built for broad, responsible public deployment. Anthropic engineered Fable 5 with safeguards that allow the Mythos-class performance to ship publicly without the unrestricted access controls that restricted the original Mythos Preview.
The result is the highest-performing AI model ever made generally available — and it is accessible today on the Claude API, Amazon Bedrock, and GitHub Copilot.
Claude Fable 5 Benchmarks: Full Performance Breakdown
Claude Fable 5 is state-of-the-art on nearly every tested benchmark of AI capability. The numbers are not marginal improvements — they represent a new performance tier.
Claude Fable 5 vs Competitors: Side-by-Side
| Benchmark | Claude 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% | — | — | — | | Humanity's Last Exam | 59.0% | — | — | — | | FrontierCode Diamond | 29.3% | 13.4% | 5.7% | — |
Source: Anthropic, MangoMind benchmark analysis
SWE-Bench Pro: 80.3% — 11 Points Clear of the Field
On SWE-Bench Pro, the most rigorous software engineering benchmark in use today, Claude Fable 5 scores 80.3 percent — an 11.1-point jump over Opus 4.8's 69.2%, released just 12 days earlier, and a 21.7-point gap over GPT-5.5 at 58.6%. This is not a competitive benchmark result. It is a generational gap.
SWE-Bench Pro measures a model's ability to resolve real GitHub issues in real codebases without hints. An 80.3% score means Fable 5 autonomously closes 4 out of every 5 real-world software engineering tasks thrown at it. That is the capability class now available via API.
Terminal-Bench 2.1: 88.0%
Terminal-Bench 2.1 evaluates autonomous command-line task completion — the exact capability class that matters for agentic security workflows. An 88% score means Fable 5 completes complex multi-step terminal operations with very high reliability. For security tooling, this translates directly to autonomous vulnerability investigation, exploit chain reconstruction, and remediation execution.
Humanity's Last Exam: 59.0%
Humanity's Last Exam is a benchmark designed to defeat expert human performance by combining questions across dozens of scientific and academic domains simultaneously. A 59% score from Fable 5 places it firmly above expert human level on one of the hardest general-knowledge evaluations in existence.
FrontierCode Diamond: 29.3% — More Than Double Second Place
On FrontierCode's Diamond split — the hardest tier of a competition-level coding evaluation — Fable 5 scores 29.3%. Opus 4.8 scores 13.4%. GPT-5.5 scores 5.7%. Fable 5 is more than twice as capable as the second-best model and more than five times as capable as the third. For security teams evaluating AI for code analysis, this gap is the relevant signal.
Claude Fable 5 Capabilities: What It Actually Does
Fable 5 is state-of-the-art across multiple capability domains simultaneously:
Software Engineering
The 80.3% SWE-Bench Pro score is the headline, but the practical capability extends well beyond benchmark tasks. Fable 5 can review complex multi-file codebases, identify architectural vulnerabilities, generate complete feature implementations, and resolve real GitHub issues autonomously. For security teams, this means AI-powered code analysis at a level that catches what senior engineers catch — at machine speed.
Knowledge Work and Reasoning
Fable 5 performs exceptionally on knowledge work tasks — research synthesis, technical documentation, policy analysis, and multi-step reasoning across domains. The Humanity's Last Exam score of 59% reflects a model that handles expert-level questions across physics, biology, law, and mathematics simultaneously.
Vision and Multimodal Analysis
Fable 5 includes advanced vision capabilities — it can analyze architectural diagrams, network topology screenshots, code screenshots, and security scan outputs. For security workflows that involve visual analysis of infrastructure or application state, vision is a first-class capability. Anthropic's multimodal research underpins this capability.
Scientific Research
Anthropic highlights scientific research as an area of exceptional Fable 5 performance. Vulnerability research, threat intelligence synthesis, and security advisory analysis all benefit from a model with genuine scientific reasoning capability rather than pattern-matching on training data.
Agentic Task Completion
The 88% Terminal-Bench 2.1 score is the agentic capability marker. Fable 5 can operate autonomously in multi-step workflows — tool calling, command execution, result evaluation, and iteration — with reliability high enough to deploy in production agentic systems. Anthropic's model spec and agentic design principles underpin this reliability. This is the capability class that powers meaningful security automation, not just assisted analysis.
Claude Fable 5 Availability and Pricing
Release date: June 9, 2026
Available on:
- Claude API (direct access)
- Amazon Bedrock — announced by Amazon
- GitHub Copilot (generally available as of June 9)
- Claude.ai (Pro, Max, Team, and Enterprise plans)
Pricing:
- $10 per million input tokens
- $50 per million output tokens
- Free at no additional cost for Pro, Max, Team, and Enterprise plan subscribers through June 22, 2026
Pricing context: Fable 5 is priced at double Opus 4.8 ($5/$25) and at half the price of Claude Mythos Preview. You get Mythos-class capability at half the Mythos Preview price — a meaningful cost efficiency improvement for organizations that were already evaluating Mythos access.
How Anthropic Made Mythos-Class AI Safe for Public Release
The original Claude Mythos Preview was restricted to Project Glasswing partners because its raw offensive security capabilities — 73% on expert hacking tasks, unlimited reverse engineering access — required controlled deployment infrastructure that most organizations do not have. We covered the full Mythos story in our Anthropic Mythos release deep-dive.
Fable 5 solves this with a capability routing system: requests that fall into specific high-risk categories — cybersecurity offense, biology, chemistry, and model distillation — are automatically routed to the less capable Claude Opus 4.8 instead of Fable 5. Anthropic says these fallbacks occur in fewer than 5 percent of sessions, meaning 95%+ of real-world usage gets full Fable 5 capability.
The design is deliberate: Mythos-class performance across software engineering, knowledge work, vision, and scientific research — with specific capability gates on the domains that required ASL-3 safeguards. You get the capability that matters for 95% of professional use cases at full Fable 5 performance. The 5% that would require unrestricted Mythos access routes to Opus 4.8, not an error.
This is what responsible frontier model deployment looks like at scale. Anthropic's Responsible Scaling Policy required ASL-3 safeguards before public release — Fable 5 is the result of that constraint done right.
Why Claude Fable 5 Matters for Security Programs in 2026
The Capability Your Threat Actors Now Have
Fable 5's SWE-Bench Pro score means threat actors now have access to an AI that can autonomously resolve real-world GitHub issues at 80% success rate. Code analysis, exploit development assistance, and vulnerability research that previously required senior engineers with specialized knowledge is now automated. The attack surface calculation for your software has changed again — and it changed today. NBC News described Fable 5 as "built on the same tech that spooked the government."
The Capability Your Defenders Now Have
The same performance that benefits threat actors is available to defenders. Fable 5's 29.3% FrontierCode Diamond score means AI-powered code security review now catches vulnerability classes that no previous publicly available model could reliably identify. Your security team, augmented by Claude Fable 5, operates at a capability level that outpaces what was available to well-funded adversaries a quarter ago.
Agentic Security at 88% Terminal Reliability
Terminal-Bench 2.1 at 88% is the number that determines whether agentic security workflows run reliably in production. At 88%, Fable 5 completes multi-step autonomous security tasks — vulnerability investigation, exploit chain reconstruction, remediation execution — with failure rates low enough to deploy in real programs. TechCrunch notes this is the line between demo-quality agentic AI and production-quality agentic AI.
Safeguard Has Already Integrated Claude Fable 5
Safeguard has integrated Claude Fable 5 natively into the Multi-Agent TAOR Deep Think AI Engine. This integration is live today for all customers — not on a roadmap, not in beta.
Here is what Fable 5 integration means across the Safeguard platform:
Fable 5-Powered Zero-Day Discovery
Fable 5's FrontierCode Diamond performance (29.3%) directly translates to vulnerability discovery depth. Safeguard routes zero-day discovery workflows through Fable 5 as the primary analysis agent, with findings verified by independent agents before surfacing. You get Fable 5's detection capability — identifying vulnerability classes that Opus 4.8 misses — within a verification architecture that eliminates the false positive burden of running a raw frontier model.
Mythos-Class Analysis at Production Scale
Because Fable 5 is priced at half the Mythos Preview rate, Safeguard can run Mythos-class analysis continuously across your full dependency graph at a cost structure that works for production security programs. Continuous zero-day discovery with Mythos-class capability is no longer reserved for organizations that can afford Mythos Preview pricing.
Agentic Remediation With Fable 5's 88% Terminal Reliability
The 88% Terminal-Bench 2.1 score means Safeguard's remediation agents, powered by Claude Fable 5, complete autonomous fix workflows with production-grade reliability. When Safeguard's discovery pipeline surfaces a confirmed vulnerability, Fable 5 remediation agents generate context-specific fix guidance, create pull requests directly in your codebase, and verify that fixes do not introduce regressions — autonomously, without human engineering time in the loop.
Agentic AI Native: Fable 5 as an Agent, Not an API Call
Safeguard is not a wrapper around Fable 5. It is an agentic AI system where Claude Fable 5 operates as an agent — alongside verification agents, supply chain intelligence agents, and remediation agents — with shared context, structured tool access, and policy-governed execution. The 88% terminal task reliability and 80% SWE-Bench performance compound in a multi-agent architecture. Each agent operates at Fable 5 capability level. The system as a whole exceeds what any single agent achieves alone.
Full Supply Chain Coverage With Vision Analysis
Fable 5's vision capability integrates with Safeguard's supply chain intelligence to bring visual analysis into the security workflow — network topology review, infrastructure diagram analysis, dependency graph visualization, and security scan output interpretation. Supply chain findings surface with full visual context, not just raw data.
Continuous Discovery, Not Quarterly Scans
Claude Fable 5 runs against your dependency graph on every code change. New packages, updated dependencies, changed code paths — all in scope the moment they land. With Fable 5's SWE-Bench-class code analysis capability running continuously, zero-day discovery is a property of your development pipeline, not a scheduled engagement.
Claude Fable 5 vs Claude Mythos Preview: Which Do You Need?
| | Claude Fable 5 | Claude Mythos Preview | |---|---|---| | Public availability | General availability | Project Glasswing only | | Pricing | $10/$50 per M tokens | $20/$100 per M tokens (est.) | | Cybersecurity offense | Routed to Opus 4.8 (<5% sessions) | Full capability | | SWE-Bench Pro | 80.3% | Comparable or higher | | Agentic use cases | Full capability | Full capability | | Best for | Production programs, continuous coverage | Specialized offensive security research |
For the vast majority of security programs — continuous zero-day discovery, supply chain coverage, agentic remediation, code security review — Fable 5 is the right model. It delivers Mythos-class performance at half the price, across 95%+ of the use cases that matter, with the ASL-3 safeguards that allow responsible deployment at scale.
Frequently Asked Questions About Claude Fable 5
What is Claude Fable 5? Claude Fable 5 is Anthropic's most capable publicly available AI model, released June 9, 2026. It is the first model from Anthropic's Mythos capability tier to be made generally available, featuring state-of-the-art performance across software engineering, knowledge work, vision, and scientific research.
When was Claude Fable 5 released? Claude Fable 5 was released on June 9, 2026, and is now generally available on the Claude API, Amazon Bedrock, and GitHub Copilot.
What is Claude Fable 5's SWE-Bench Pro score? Claude Fable 5 scores 80.3% on SWE-Bench Pro — 11.1 points ahead of Opus 4.8 (69.2%), 21.7 points ahead of GPT-5.5 (58.6%), and 26.1 points ahead of Gemini 3.1 Pro (54.2%). Source: Anthropic.
How does Claude Fable 5 compare to Claude Mythos Preview? Claude Fable 5 is the publicly available version of Mythos-class capability. It delivers equivalent performance across most use cases at half the price, with capability routing that safely manages the subset of high-risk requests (fewer than 5% of sessions) that required restricted access in the original Mythos Preview.
What is Claude Fable 5's pricing? Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens. It is available at no additional cost through June 22, 2026 for Claude Pro, Max, Team, and Enterprise subscribers.
What safeguards does Claude Fable 5 have? Anthropic routes specific high-risk request categories — cybersecurity offense, biology, chemistry, and model distillation — to Claude Opus 4.8 instead of Fable 5. This occurs in fewer than 5% of sessions. The routing enables broad public release of Mythos-class capability while managing the specific domains that required ASL-3 restricted access.
Is Claude Fable 5 available on Amazon Bedrock? Yes. Claude Fable 5 is available on Amazon Bedrock as of its June 9, 2026 release date.
Is Claude Fable 5 available on GitHub Copilot? Yes. Claude Fable 5 is generally available for GitHub Copilot as of June 9, 2026.
How has Safeguard integrated Claude Fable 5? Safeguard has integrated Claude Fable 5 natively into its Multi-Agent TAOR Deep Think AI Engine as a first-class agent for zero-day discovery, vulnerability remediation, and supply chain security. The integration is live for all customers as of today. Fable 5 operates as an agent within Safeguard's multi-agent architecture — not as an API call — delivering Mythos-class security analysis continuously across customer dependency graphs.
What is the difference between Claude Fable 5 and Claude Mythos? Claude Mythos Preview was restricted to approximately 50 Project Glasswing defensive security partners from April to June 2026 due to its unrestricted offensive security capabilities. Claude Fable 5 is the Mythos-class model built for general availability — equivalent performance across the vast majority of use cases, with capability routing for the high-risk domains that required restricted access in Mythos Preview. Read our full Anthropic Mythos breakdown here.
The New Baseline
Anthropic just made its most capable AI model available to everyone. The performance gaps over GPT-5.5 and Gemini 3.1 Pro are not competitive advantages measured in decimal points — they are structural gaps that reflect a different capability tier. At 80.3% on SWE-Bench Pro, 88% on Terminal-Bench 2.1, and 29.3% on FrontierCode Diamond, Claude Fable 5 is in a class by itself among publicly available models.
For security programs, the question is not whether Fable 5 changes your threat model. It does — the same capability that is now on your side is now on the other side. The question is how fast you operationalize it.
Safeguard integrated Fable 5 before the public launch. Mythos-class zero-day discovery, Fable 5-powered remediation, and continuous agentic security coverage are live today — not coming soon. If you want Fable 5's capability running on your dependency graph with verification, remediation, and supply chain context built in, we are already there.
Safeguard's Claude Fable 5 integration is live for all customers. To see Fable 5-powered zero-day discovery running on your actual stack — with the full agentic remediation pipeline behind it — reach out. We will run a live demonstration against your dependency graph.
Sources:
- Claude Fable 5 and Claude Mythos 5 — Anthropic Official Announcement
- Anthropic releases Mythos-like AI model to the public, Claude Fable 5 — CNBC
- Claude Fable 5 Benchmarks: 80.3% SWE-Bench Pro, 11 Points Ahead of the Field
- Claude Fable 5 Benchmarks Full Analysis — MangoMind
- Claude Fable 5 from Anthropic now available on Amazon Bedrock
- Claude Fable 5 is generally available for GitHub Copilot — GitHub Changelog
- Anthropic releases Fable 5 model, built on the same tech that spooked the government — NBC News
- Anthropic's Claude Fable 5 is a version of Mythos the public can access today — TechCrunch
- Anthropic Responsible Scaling Policy
- SWE-Bench: Can Language Models Resolve Real-World GitHub Issues?
- Humanity's Last Exam Leaderboard — Scale AI