TL;DR Claude Fable 5 is Anthropic's newest generally available frontier model, built on the same foundation as Mythos 5 with additional safeguards for cybersecurity and biology-related queries. It sets new highs on benchmarks like SWE-Bench Pro (80.3%), outperforms Opus 4.8 and GPT-5.5 across many coding and agent tasks, and introduces a safeguard system that falls back to Opus 4.8 for sensitive requests. This guide breaks down its benchmarks, pricing, safeguards, and where it fits in production AI stacks.

Claude Fable 5 launched on June 9, 2026, and it's the most capable model Anthropic has made generally available. It's built on the same underlying architecture as Mythos 5, the frontier model currently restricted to a small set of organisations through Project Glasswing, but with a safety layer added for broader release. That distinction matters, and we'll get into it in detail below.

This post covers what Claude Fable 5 is, how it performs across the benchmarks that actually matter for production use, what the safeguard system does in practice, and where it stands against Opus 4.8 and GPT-5.5.

Fable 5 is the most capable generally available model Anthropic has ever shipped. Want to see what generative UI looks like with it?
Try out Thesys Agent Builder that lets you build agents that respond with adaptive visual interfaces instead of plain text.

Build Your AI Agent with Fable 5

What Is Claude Fable 5?

Claude Fable 5 is Anthropic's new general-availability frontier model. It runs on the same weights as Mythos 5 but ships with safeguards enabled, primarily covering cybersecurity and biology-related queries. Outside those categories, performance is equivalent to Mythos 5.

The safeguards trigger in fewer than 5% of sessions on average. When they do, affected queries fall back to Claude Opus 4.8. Anthropic has described the current tuning as conservative, with plans to reduce false positives over time.

For most developers and teams, that means Fable 5 behaves like Mythos 5 across the vast majority of tasks, making it a strong candidate for production-grade agent deployments. The 5% figure is an average across all use cases; teams working in sensitive domains should expect a higher rate. For teams evaluating which model to build agents around, the benchmark gaps are becoming increasingly meaningful.

Claude Fable 5 Benchmarks: How Does It Compare?

Here's the full benchmark table across Fable 5/Mythos 5, Mythos Preview, Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.

Benchmark	Fable 5 / Mythos 5	Mythos Preview	Opus 4.8	GPT-5.5	Gemini 3.1 Pro
SWE-Bench Pro	80.3%	77.8%	69.2%	58.6%	54.2%
FrontierCode Diamond	29.3% (xhigh)	—	13.4%	5.7%	—
GDPval-AA (Elo)	1932	—	1890	1769	1314
GDPpdf	29.8%	—	22.5%	24.9%	16.7%
Blueprint-Bench 2	38.6%	—	14.5%	36.2%	26.5%
AutomationBench	17.4%	—	15.5%	12.9%	9.6%
OSWorld-Verified	85.0%	85.4%	83.4%	78.7%	76.2%
Legal Agent Benchmark	13.3%	—	10.4%	2.1%	0.0%
HLE (no tools)	59.0%*	56.8%	49.8%	41.4%	44.4%
HLE (with tools)	64.5%*	64.7%	57.9%	52.2%	51.4%
BioMysteryBench (hard)	46.1%*	29.6%	40.0%	—	—
BioMysteryBench (human solved)	83.9%*	82.6%	80.4%	—	—
Terminal-Bench 2.1	88.0%*	—	82.7%	83.4%	70.7%
ExploitBench (cybersecurity)	78.0%*	69.0%	40.0%	34.0%	—
HealthBench Professional	66.0%*	64.7%	56.9%	51.8%	—

*Starred benchmarks reflect Mythos 5 scores. Because Fable 5's safeguards cover cybersecurity and biology, queries in those categories fall back to Opus 4.8 in production. The starred numbers show what the underlying model can do; real-world performance on those tasks is closer to Opus 4.8's figures.

Where Fable 5 Leads Clearly

Coding: SWE-Bench Pro at 80.3% is a 11.1-point improvement over Opus 4.8, Anthropic's previous flagship model (69.2%) and a 21.7-point gap over GPT-5.5 (58.6%). FrontierCode Diamond, which tests the hardest tier of coding problems, comes in at 29.3% for Fable 5 versus 13.4% for Opus 4.8 and 5.7% for GPT-5.5. That's more than double Opus 4.8 on the hardest coding tasks.

Legal reasoning: The Legal Agent Benchmark result is the starkest competitive gap in the table. Fable 5 scores 13.3%. GPT-5.5 scores 2.1%. Gemini 3.1 Pro scores 0.0%. Legal reasoning at agent scale is not a close race here.

Spatial and document reasoning: Blueprint-Bench 2 at 38.6% beats GPT-5.5 (36.2%) and is more than double Opus 4.8 (14.5%). GDPpdf at 29.8% outperforms all listed competitors on document understanding.

Agent tasks: Agent tasks: AutomationBench at 17.4% leads GPT-5.5 (12.9%) and Gemini (9.6%), a meaningful signal for teams building long-running autonomous workflows.

The Starred Benchmark Context

For cybersecurity and biology benchmarks, the numbers in the table represent Mythos 5's capability, not what Fable 5 will do in a typical deployment. ExploitBench shows 78.0% for Mythos 5, but Fable 5 falls back to Opus 4.8 on those queries, which scores 40.0%.

The same applies to BioMysteryBench, HealthBench Professional, and Terminal-Bench 2.1. If your use case touches those domains regularly, the starred figures won't reflect your experience.

Claude Fable 5 vs. Opus 4.8: Is It Worth Switching?

For most production workloads, yes. The performance gap is large enough on coding, document reasoning, and agent tasks that the upgrade is straightforward to justify.

The main consideration is the safeguard fallback. If your application handles cybersecurity tooling, security research, or biology-adjacent content, some proportion of your queries will route to Opus 4.8 rather than Fable 5. That's not necessarily a dealbreaker, but it's worth factoring in before switching.

For everything else, including long-horizon coding tasks, financial analysis, legal reasoning, and multimodal document work, Fable 5 is a meaningful step up.

Real-world evidence supports this. Stripe used the underlying Mythos 5 model to complete a codebase-wide migration of a 50-million-line Ruby codebase in a single day, a task that had previously taken a full engineering team two months. Cursor called it "state of the art on CursorBench" and noted it opened up "long-horizon problems out of reach for earlier models." GitHub highlighted its ability to handle "complex long-horizon coding tasks with autonomy and reliability that exceeded previous benchmarks."

The through-line across these cases: the longer and more complex the task, the larger Fable 5's advantage tends to be.

Claude Fable 5 vs. GPT-5.5: A Brief Comparison

Fable 5 leads OpenAI's latest reasoning model on most of the benchmarks listed above. The gap is widest on SWE-Bench Pro (21.7 points), FrontierCode Diamond (23.6 points), and the Legal Agent Benchmark (11.2 points). On GDPval-AA, the Elo gap is 163 points. On Blueprint-Bench 2, Fable 5 edges ahead 38.6% to 36.2%.

GPT-5.5 closes the gap on HLE without tools (41.4% vs. 44.4% for Gemini, but Fable 5 still leads at 59.0%) and performs competitively on OSWorld-Verified (78.7% to Fable 5's 85.0%).

For finance-specific work, Fable 5 achieved the highest score of any model on the Hebbia Finance Benchmark, which tests senior-level financial reasoning. IMC found it scored top marks across factual lookup, conceptual reasoning, root cause analysis, and expected-value analysis in trading evaluations.

Claude Fable 5 Pricing

Fable 5 is priced at $10 per million input tokens and $50 per million output tokens.

That's less than half the price of Mythos Preview. For teams that have been waiting on Mythos Preview's pricing to make frontier-level capability viable at scale, Fable 5's general availability changes what's buildable.

The Safeguard System: What It Is and When It Triggers

The safeguards in Fable 5 are a content layer that sits on top of the Mythos 5 base model. They're primarily designed to restrict outputs in cybersecurity and biology domains where the underlying model's capability could cause harm.

When a query triggers the safeguards, it doesn't fail. It falls back to Opus 4.8 and completes normally. The average trigger rate across all session types is under 5%. Anthropic has acknowledged the current calibration is conservative and has committed to reducing false positives.

For most developer use cases, this system is invisible. For teams building security-focused tooling or applications in life sciences, it's worth testing your specific queries to understand your actual fallback rate before committing to a migration.

What About Project Glasswing?

Project Glasswing is the programme through which Anthropic has been making Mythos Preview available to a small set of vetted organisations. It's not a public offering.

Fable 5 is the generally available version of the same underlying model, with safeguards applied. The two coexist: Mythos Preview remains available through Glasswing for organisations whose use cases require unconstrained access and have passed Anthropic's vetting process. Fable 5 is the path for everyone else.

Build with Fable 5 on Agent Builder

Fable 5 is available now in Thesys Agent Builder. Agent Builder lets you build AI systems that can reason, plan, and take action powered by Fable 5 that respond with interactive charts, tables, and visual components, not plain text. If you're building anything that involves complex reasoning over documents, financial data, or code, it's worth trying with Fable 5 as the backbone.

Build Your AI Agent Today

FAQs: Claude Fable 5

What is Claude Fable 5?

Claude Fable 5 is Anthropic's latest generally available model, launched June 9, 2026. It's built on the same base as Mythos 5 but includes safety safeguards that restrict outputs in specific sensitive domains, primarily cybersecurity and biology. Outside those categories, performance is equivalent to Mythos 5, which currently leads or matches every publicly available model on most benchmarks.

How does Fable 5 differ from Mythos 5?

Fable 5 and Mythos 5 share the same underlying model weights. The difference is that Fable 5 ships with a safeguard layer that triggers on certain query types. When those safeguards activate, which happens in under 5% of sessions on average, the query falls back to Opus 4.8. Mythos 5, available as Mythos Preview through Project Glasswing, has no such restrictions but is not generally available.

How does Claude Fable 5 compare to Opus 4.8?

Fable 5 outperforms Opus 4.8 by a significant margin on coding, legal reasoning, document understanding, and agent tasks. On SWE-Bench Pro, Fable 5 scores 80.3% versus Opus 4.8's 69.2%. On FrontierCode Diamond, Fable 5 scores 29.3% versus 13.4%. On cybersecurity and biology benchmarks, where Fable 5 falls back to Opus 4.8 due to safeguards, the gap narrows or disappears.

How does Claude Fable 5 compare to GPT-5.5?

Fable 5 leads GPT-5.5 on most benchmarks, with the largest gaps on SWE-Bench Pro, FrontierCode Diamond, and the Legal Agent Benchmark. GPT-5.5 performs more competitively on general knowledge tasks but doesn't close the gap on coding or reasoning benchmarks.

What is the Claude Fable 5 pricing?

Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens. This is less than half the price of Mythos Preview.

What are the safeguards in Fable 5 and when do they trigger?

The safeguards are a content-filtering layer designed primarily for cybersecurity and biology-adjacent queries. When they trigger, the query falls back to Opus 4.8 rather than failing. The average trigger rate is under 5% of sessions. Anthropic has said the current calibration is conservative and they plan to reduce false positives over time.

Is Claude Fable 5 available now?

Yes. Claude Fable 5 launched on June 9, 2026, and is generally available.

What is Project Glasswing?

Project Glasswing is Anthropic's programme for making Mythos Preview, the unconstrained version of the Mythos 5 model, available to a small number of vetted organisations. It is not a public offering. Fable 5 is the generally available alternative for teams that don't need unconstrained frontier access.

Related articles

May 2026 Security Incident

Why We're Open Sourcing OpenUI

AI Copilots: A Complete Guide to Intelligent Assistance

What Are AI Agents: Types, Capabilities, and Real-World Use

ChatKit: What It Is, What It Isn’t, and Where C1 Fits In

PowerPoint Automation Reimagined with C1 Artifacts API

The Future of AI Interfaces is Generative

MCP UI: Making AI Apps Interactive

What Is AG-UI and Why It Matters for AI-Native User Interfaces

How to design AI-Native Conversational Interfaces : From Templates to Generative UI

GPT 5 vs. GPT 4.1

How to Build Generative UI Applications

Implementing Generative Analytics with Thesys and MCP

Evolution of Analytics: From Static Dashboards to Generative UI

Why Generating Code for Generative UI is a bad idea

Building the First Generative UI API: Technical Architecture and Design Decisions Behind C1

How we evaluate LLMs for Generative UI

Generative UI vs Prompt to UI vs Prompt to Design

Generative UI: What It Is and Why It Matters