Claude Fable 5 Is Here: What the New Top Model Means for Your Daily AI Work
Anthropic released Claude Fable 5 on June 9, 2026 — its most capable publicly available model, with major gains in long-horizon tasks, coding, document reasoning, and vision. It's free on Pro, Max, and Team plans through June 22. Here's what changed and whether you should switch.
TL;DR. Anthropic released Claude Fable 5 on June 9, 2026 — a Mythos-class model that outperforms every previous Claude on coding, document reasoning, and long-horizon tasks. It's free on Pro, Max, Team, and Enterprise plans through June 22; after that, usage credits apply. It counts as roughly 2× usage on subscription plans. Via API:
claude-fable-5at $10/$50 per million tokens.
Anthropic shipped two models on June 9, 2026: Claude Fable 5, now publicly available, and Claude Mythos 5, restricted to vetted research partners. For most professionals using Claude.ai or the API, Fable 5 is what matters — and the most important thing to know right now is that you can try it free until June 22.
Here's what changed, what it's actually better at, and how to think about switching.
What "Mythos-class" means
Anthropic's model tiers have historically been Haiku (fast/light), Sonnet (balanced), and Opus (most capable). Fable 5 introduces a new category above Opus: Mythos-class. Think of it as the jump from "top of the existing line" to a new tier — not an incremental Opus update, but a substantially different model.
The benchmark numbers make this concrete. On SWE-Bench Pro — a test of real-world software engineering tasks — Fable 5 scores 80.3% against Opus 4.8's 69.2%. On FrontierCode, it scores 29.3% against GPT 5.5's 5.7% and Opus 4.8's 13.4%. On Hebbia's Finance Benchmark, which tests senior-analyst-level reasoning over complex documents, Fable 5 tops the leaderboard.
These aren't cherry-picked single-domain wins. The model is better across the domains professionals actually care about: long-form reasoning, document analysis, coding, and vision.
What's actually different for the work you do
The capability change that matters most for professional use isn't the benchmark number — it's how long the model can hold a task together.
Sustained task execution. Fable 5 can work through multi-step workflows for hours — Anthropic's product page describes "days at a time" in agentic configurations, and researcher Ethan Mollick described sessions running "up to a dozen hours" on multi-page specifications. For most daily professional use, this translates to: fewer breakdowns mid-task, better coherence in long documents, and less re-prompting when you're doing something genuinely complex.
Stripe reported compressing five months of engineering work into days using Fable 5 in an agentic pipeline. That's an enterprise example, but the underlying capability — not losing the thread in step 8 of a 10-step task — shows up in smaller ways for individual professional users too.
Document and financial reasoning. The model tops Hebbia's Finance Benchmark, which tests performance on tasks like reading dense regulatory filings, cross-referencing contract terms, and synthesizing multi-document analysis. For attorneys reviewing complex agreements, financial advisors building reports from multiple data sources, or compliance officers working through regulatory updates, this is the practical improvement to look for.
Vision. Fable 5 sets a new benchmark for vision tasks. In testing it has rebuilt complete web applications from nothing more than a screenshot — which is impressive as a demo, but the practical version is: it reads charts, tables, medical images, and embedded figures in PDFs more reliably than Opus 4.8.
Code. The coding gains are significant enough that Claude Code users should notice them immediately. Fable 5 writes its own tests, handles large migrations autonomously, and recovers from mid-task errors without losing context. For professionals running Python-light analysis, automating document workflows, or using Claude Code for professional tasks, this is the biggest day-to-day improvement.
What Mythos 5 is (and why most readers won't access it)
Claude Mythos 5 is the more capable sibling — restricted via Anthropic's Project Glasswing program to vetted cybersecurity researchers and select biomedical labs. It scored 78% on ExploitBench for cybersecurity tasks and matched experienced human operators on drug-design workflows.
If you're reading this as a professional using Claude.ai or the API for business work, Mythos 5 is not available to you, and that's intentional. It's not behind a paywall — it's behind a vetting and safety review process designed to limit access to the highest-risk capabilities. Fable 5 is the model that delivers the practical benefits of the Mythos generation to the rest of us.
A note on the safety architecture: when Fable 5 receives queries in restricted domains (offensive cybersecurity, certain biology/chemistry research, model distillation attempts), it routes to Claude Opus 4.8 rather than refusing outright. This affects fewer than 5% of sessions in practice, and the vast majority of professional users will never see it.
The subscription situation right now
The window to use Fable 5 without managing usage credits is short.
Through June 22, 2026: Fable 5 is included in Claude.ai Pro, Max, Team, and Enterprise plans at no extra cost. Enable it by clicking the model selector at the top of the Claude.ai chat window and choosing Claude Fable 5.
After June 22: Anthropic will move Fable 5 to a usage-credit model while they scale capacity, with the stated intention to restore it as a standard subscription feature "as soon as possible." The exact credits structure hasn't been announced. Max plan subscribers, who already have the deepest usage pool, will be best-positioned to absorb the cost.
On API: Available now at claude-fable-5. Priced at $10/million input tokens and $50/million output tokens — double Opus 4.8's rate. Also available on AWS Bedrock, Google Cloud Vertex AI, Microsoft Azure AI Foundry, and as of June 9, GitHub Copilot.
Should you switch now?
For quick, single-turn tasks — short drafts, direct questions, simple summaries — Opus 4.8 and Sonnet 4.6 remain fast and efficient. Fable 5 is worth the extra usage when the task genuinely benefits from sustained reasoning: long-document analysis, multi-step workflows, complex code, or any session where you'd otherwise be doing significant re-prompting.
Given that Fable 5 is free on your subscription through June 22, the answer to "should I try it?" is yes — try it now, use it on a task that represents your most complex actual work, and you'll have a calibrated sense of the improvement before the credit model kicks in.
One constraint to flag: Anthropic requires 30-day data retention for Fable 5 sessions for safety monitoring purposes. If your organization has a zero-retention data policy for AI — common in certain legal, healthcare, and financial contexts — check whether that requirement changes your deployment plans before switching.
Sources
- Anthropic: Introducing Claude Fable 5 and Mythos 5
- Anthropic product page: Claude Fable 5
- The Decoder: Anthropic releases Claude Fable 5 and Mythos 5 with major gains in coding and science
- TechCrunch: Anthropic's Fable 5 can make weirdly fun video games with the click of a button
Save hours every week with the AI Career Lab — All AI Prompts Bundle
All eight profession-specific AI Prompts packs — 393 agentic skills total with ambient compliance guards. Runs on Claude Cowork.
Frequently asked questions
Is Claude Fable 5 available on my Claude.ai subscription right now?+
Yes — as of June 9, 2026, Fable 5 is available on Claude.ai Pro, Max, Team, and Enterprise plans at no extra cost through June 22, 2026. After that date, it will require usage credits while Anthropic scales capacity, with plans to restore it as a standard subscription feature. To access it now, open Claude.ai, click the model selector near the message box, and choose Claude Fable 5. Note that on subscription plans, Fable 5 counts as approximately 2× usage relative to Opus 4.8.
What is Claude Fable 5, and how does it compare to Claude Opus 4.7?+
Claude Fable 5 is Anthropic's new top publicly available model, released June 9, 2026. It represents a new generation — Mythos-class — that sits above the Opus tier. On coding benchmarks, it scores 80.3% on SWE-Bench Pro versus 69.2% for Opus 4.8, and 29.3% on FrontierCode versus 13.4% for Opus 4.8. It's meaningfully better at long multi-step tasks, document-heavy reasoning, and vision analysis. The short version: Fable 5 can hold context and execute tasks across hours-long sessions in ways Opus 4.7 could not.
What is Claude Mythos 5, and can I use it?+
Claude Mythos 5 is a more capable but restricted version of the same generation. It's available only to vetted partners through Anthropic's Project Glasswing program — primarily cybersecurity researchers and select biomedical labs. Most Claude.ai subscribers and API users will not have direct access to Mythos 5. For professional work, Fable 5 is the relevant new model.
Does Claude Fable 5 change my data privacy settings?+
Anthropic requires 30-day data retention for Fable 5 safety monitoring — this is a change from zero-retention policies some API users previously maintained. Enterprise teams with strict data-handling requirements should review Anthropic's updated data terms before deploying Fable 5 in production workflows.
When should I stick with Opus 4.8 or Sonnet 4.6 instead of switching to Fable 5?+
Fable 5 has a higher usage cost (roughly 2× credits on subscriptions; $10/$50 per million tokens via API versus $5/$25 for Opus 4.8). For quick single-turn tasks — one-off questions, short drafts, simple summaries — Opus 4.8 and Sonnet 4.6 remain solid and cheaper. Switch to Fable 5 when the task genuinely benefits from deeper reasoning: complex document analysis, multi-step workflows, code reviews, or any session where you're expecting to iterate for more than a few exchanges.
Related Guides
Apple's New Siri Runs on Google AI. Does That Put Your Work Data at Risk?
Apple rebuilt Siri from scratch on a custom Google Gemini model at WWDC 2026. Here's exactly how the three-tier privacy architecture works, what the contract with Google prevents, and whether it's safe to use for work.
OpenAI Says 'Chat Is Dead': What ChatGPT's Agent App Pivot Means for You (2026)
OpenAI is rebuilding ChatGPT from a chatbot into a full agent app — able to handle multi-step tasks like booking travel, writing code, and managing projects with minimal input. Here's what's changing, when, and what it means if you use ChatGPT for work.
ChatGPT Just Added a Data-Safety Switch: What Lockdown Mode Does and Who Needs It
OpenAI rolled out Lockdown Mode on June 6, 2026 — an optional security toggle that blocks the most common way sensitive data can leak through prompt injection attacks. Here's what it is, what you lose, and whether to turn it on.