Anthropic just released Claude Fable 5 on June 9, 2026 — the most powerful AI model the company has ever made available to the public.
For the first time, the same Mythos-class intelligence previously locked behind a closed partner program is in your hands.
In this guide: what actually changed, how it compares to Opus 4.8, what the pricing looks like, and the smartest ways to put it to work immediately.
- 80.3% — 11 points ahead of the next model
- 50M-line Ruby codebase migrated in 1 day
- Top score on Cognition's FrontierCode eval
- First place on Hebbia's senior analyst evaluation
- First model above 90% on Hex analytics test
- Strong across factual lookup and root-cause analysis
- 5× larger than Claude Opus 4.8 (200K tokens)
- Up to 128,000 output tokens per response
- Full archives, codebases, or strategies in one session
Some links are affiliate links · You support us at no extra cost · Full disclosure
What Is Claude Fable 5?
Claude Fable 5 is Anthropic's first publicly available Mythos-class AI model, released June 9, 2026. It represents a new tier above the Opus family — state-of-the-art on nearly every benchmark.
It ships with a 1-million-token context window and handles complex, multi-step tasks autonomously over hours or days.
Most AI model launches are version bumps dressed in a press release. Claude Fable 5 is not that.
The Mythos-class designation matters. Until June 9, this level of intelligence only existed inside Project Glasswing — a restricted program involving AWS, Microsoft, Apple, and CrowdStrike.
Anthropic opened that capability to the public for the first time. But it ships with one important caveat: Claude Fable 5 includes safety classifiers that reroute certain high-risk queries to Claude Opus 4.8 instead.
Claude Fable 5 and Claude Mythos 5 share the same underlying model.
Fable 5 has safety classifiers active. Mythos 5 has them partially lifted — and is only available to vetted Project Glasswing partners.
Those classifiers catch genuinely dangerous requests — advanced exploit development, high-risk biology, model distillation. They trigger in fewer than 5% of sessions.
For content creators, marketers, and most business professionals, this is completely invisible. You work with Claude Fable 5 as if the classifiers don't exist — because for 95%+ of tasks, they don't apply.
The 1-million-token context window is the clearest practical leap. Feed Claude Fable 5 an entire codebase, research archive, or document library — it holds all of it in memory at once.
We covered the previous generation in detail if you want the full version timeline: Claude Opus 4.6: what changed and who should use it.
Claude Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans through June 22, 2026.
After June 23, using it requires usage credits. Anthropic's goal is to restore it as a standard plan feature as capacity grows.
Claude Fable 5 vs Opus 4.8: What Actually Changed
Here's the direct comparison. Note that Opus 4.8 still wins on output price and plan availability — Claude Fable 5 earns its premium entirely on capability and benchmark performance.
| Feature | Claude Fable 5 | Claude Opus 4.8 | Winner |
|---|---|---|---|
| Use Case | General use (with safety classifiers) | General use | Tie |
| Benchmarks | State-of-the-art — SWE-Bench Pro: 80.3% | Solid — 11+ points behind on SWE-Bench | Fable 5 |
| Coding | 50M-line migration in 1 day | 2+ months for a full engineering team | Fable 5 |
| Context Window | 1 million tokens | 200K tokens | Fable 5 |
| Input Price | $10 / million tokens | $10 / million tokens | Tie |
| Output Price | $50 / million tokens | $25 / million tokens | Opus 4.8 |
| Plan Availability | API + Pro/Max/Team (free until June 22) | Standard on all plans — always available | Opus 4.8 |
For more context on how capability has evolved across Claude generations, see our Claude 4 generations comparison.
1. Coding — 50 Million Lines in a Day
On Cognition's FrontierCode evaluation, Claude Fable 5 leads all frontier models for production-quality output — even at medium effort settings.
The Stripe result tells the story plainly: a 50-million-line Ruby codebase migration, done by Claude Fable 5 in one day. A full team would have taken over two months.
2. Vision — Screenshots to Working Code
Claude Fable 5 can rebuild a web app's source code from screenshots alone.
Earlier Claude models needed complex helper harnesses to complete visual navigation tasks. Claude Fable 5 handles them with a minimal vision-only setup — no extra scaffolding required.
3. Knowledge Work — First at Senior Analyst Grade
Claude Fable 5 scored first place on Hebbia's Finance Benchmark for senior-level analytical reasoning.
On Hex's analytics evaluation, it crossed 90% — the first model to do so. IMC independently confirmed strong performance across factual lookup, root-cause analysis, and expected-value work.
Claude Fable 5 scores highest on coding tasks even at medium effort — you don't need to craft elaborate prompts to see the improvement.
Ready to test it? Access Claude Fable 5 here — free on Pro and Max plans until June 22.
Fable 5 vs Mythos 5: Same Engine, Different Access Level
The naming trips people up. Claude Fable 5 and Claude Mythos 5 are not two separate models. They're the same underlying system with different safety configurations applied on top.
| Claude Fable 5 | Claude Mythos 5 | |
|---|---|---|
| Underlying model | Identical | Identical |
| Safety classifiers | Active (conservative tuning) | Partially lifted |
| Available to | Everyone — API, Claude.ai, Amazon Bedrock, more | Project Glasswing partners only |
| Cybersecurity tasks | Routed to Opus 4.8 when high-risk | Full access for vetted partners |
| Pricing | $10 / $50 per million tokens | $10 / $50 per million tokens |
What Triggers the Opus 4.8 Fallback?
Four categories activate the safety reroute in Claude Fable 5:
- Advanced cybersecurity — exploit development, zero-day chaining, offensive security tooling
- High-risk biology and chemistry — dual-use research, pathogen synthesis pathways
- Model distillation — attempts to extract training data or replicate the model
- Other conservatively flagged requests — Anthropic notes the tuning errs cautious and will improve over time
When a reroute happens, Claude Fable 5 returns stop_reason: "refusal" as an HTTP 200. Refused responses are not billed if no output was generated.
For everyday business and creative work, this is a non-issue. Fewer than 5% of sessions trigger any fallback at all.
Mythos 5 is a separate deployment entirely — reserved for infrastructure providers, vetted cybersecurity researchers, and organizations approved through Project Glasswing.
Understanding where autonomous AI work is heading next is covered in our guide on AI agents and autonomous tools.
Claude Fable 5 handles virtually all business, creative, and analytical work without ever triggering the fallback.
The 5% figure covers genuinely dangerous requests — not aggressive prompting, competitor research, or security audits of your own systems.
How to Access Claude Fable 5 Right Now
There's a 12-day window where Claude Fable 5 is included at no extra cost on subscription plans. Here's exactly how to get it on every platform before June 22.
Claude Subscribers (Pro / Max / Team / Enterprise)
Claude Fable 5 is live now. Open your Claude workspace, select Fable 5 from the model picker, and start using it.
It counts as 2× your normal usage credits per message. After June 23, separate usage credits will be required until Anthropic restores it as a standard plan feature.
API Developers
Call the model with the string claude-fable-5. The API is fully available today on consumption-based Enterprise plans and standard API access.
Pricing: $10 per million input tokens and $50 per million output tokens — less than half the rate of the original Mythos Preview launch.
Claude Code, Bedrock, and GitHub Copilot
Claude Fable 5 is available in Claude Code and across third-party platforms: Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
GitHub Copilot users should note: Claude Fable 5 requires 30-day prompt-and-output retention to run its safety classifiers. This setting is off by default in Copilot admin controls.
June 9, 2026: Claude Fable 5 launches — free on Pro, Max, Team, and seat-based Enterprise.
June 23, 2026: Fable 5 removed from flat-rate plans. Usage credits required going forward.
Later 2026: Anthropic plans to restore it as a standard plan feature as capacity scales.
Full API documentation is on Anthropic's official Claude Fable page. For hands-on workflows, our Claude AI productivity guide covers 17 practical methods to apply immediately.
Best Use Cases for Solopreneurs & Digital Marketers
Ever tried to run a deep competitor analysis, a full content audit, and a strategy session — all in one afternoon?
Claude Fable 5 handles all three in sequence, without losing context between tasks.
1. Deep Research and Document Analysis
Claude Fable 5's 1-million-token context window means you can load entire research archives, contract sets, or product libraries in one go.
Ask it to extract patterns, summarize findings, or flag inconsistencies — it holds all of it at once, without forgetting earlier pages.
2. Content Strategy and Long-form Writing
On short tasks like email drafts, early testers report Claude Fable 5 feels similar to Opus 4.8.
Where the gap widens: long strategy documents, multi-angle content briefs, and posts that require synthesizing dozens of sources at once.
3. Automations and Workflow Code
You don't need to be a developer to benefit from Claude Fable 5's coding step-change.
Describe the automation you want in plain language and it writes working code — often in one attempt where Opus 4.8 needed three.
4. Financial and Business Analysis
Claude Fable 5 reached the top of Hebbia's Finance Benchmark at senior analyst grade.
For solopreneurs: feed it a P&L, a competitor pricing page, or a client contract and ask for a structured decision brief. The output quality is materially different from earlier Claude versions.
Claude Fable 5 costs 2× more per output token. For short, repetitive tasks — quick replies, simple edits, one-line lookups — Opus 4.8 delivers similar results at half the cost.
Route Claude Fable 5 toward tasks that are long, complex, or multi-step. Use Opus 4.8 for everything else.
Anthropic's system card for Fable 5 and Mythos 5 covers alignment testing in detail — worth reading if you're evaluating it for client or enterprise use.
For a broader look at building your full AI toolkit, our post on best AI tools for solopreneurs covers the complete stack from $0 to production.
Frequently Asked Questions — Claude Fable 5
What is Claude Fable 5 and when was it released?
Claude Fable 5 is Anthropic's first publicly available Mythos-class AI model, launched June 9, 2026. It sits above the entire Opus family in capability.
The API model string is claude-fable-5. It supports a 1-million-token context window and up to 128,000 output tokens per response — the largest publicly available context from Anthropic to date.
What is the difference between Claude Fable 5 and Claude Mythos 5?
Same underlying model. Different access rules. Claude Fable 5 has safety classifiers active that reroute high-risk queries to Claude Opus 4.8 in fewer than 5% of sessions.
Claude Mythos 5 has those classifiers partially lifted. It is not a general release — only available through Project Glasswing to vetted infrastructure providers and cybersecurity partners.
Claude Fable 5 is the version everyone can use. Mythos 5 is reserved for high-trust, high-stakes operators.
How much does Claude Fable 5 cost on the API?
Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens — double the output rate of Claude Opus 4.8 ($25/million).
Given the 2× output price, route Claude Fable 5 toward long, complex, multi-step tasks where the quality jump justifies the cost. Use Opus 4.8 for shorter, repetitive work.
Is Claude Fable 5 free for Claude Pro and Max subscribers?
Yes — through June 22, 2026. Claude Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans. It counts as 2× your normal usage credits.
After June 23, Anthropic removes it from flat-rate plans. Usage credits will be required until capacity expands and it returns as a standard feature.
You can try Claude Fable 5 on your existing plan here. Also check our Claude productivity guide to get the most out of your session.
Why does Claude Fable 5 sometimes respond with Claude Opus 4.8 instead?
Claude Fable 5 includes safety classifiers that reroute specific high-risk categories to Claude Opus 4.8. The four main triggers are:
Advanced cybersecurity exploit development, high-risk biology and chemistry, model distillation attempts, and a small set of conservatively flagged requests Anthropic intends to refine over time.
This happens in fewer than 5% of sessions and is invisible in the output — you receive an Opus 4.8 answer rather than an error.
Refused API calls return stop_reason: "refusal" and are not billed.
How does Claude Fable 5 compare to ChatGPT and Gemini?
Claude Fable 5 leads on agentic coding benchmarks — 80.3% on SWE-Bench Pro, 11 points ahead of the next competitor — and tops Hebbia's Finance Benchmark for senior-level document reasoning.
GPT-5.5 and Gemini 3.5 remain competitive on multimodal tasks and general reasoning. But on long-horizon coding, large-document analysis, and autonomous multi-step work, Claude Fable 5 holds a clear lead.
Can solopreneurs and digital marketers use Claude Fable 5 effectively?
Especially for complex, multi-step work. The 1-million-token context window makes Claude Fable 5 ideal for deep research, content strategy across large source sets, automation code, and financial analysis.
For short, repetitive tasks — quick edits, one-liners, simple lookups — Claude Opus 4.8 delivers similar results at half the output cost. The smart approach is routing the right task to the right model.
What is the API model string for Claude Fable 5?
The API model string is claude-fable-5. Available through the Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, GitHub Copilot, and Claude Code.
Supports up to 1 million input tokens and 128,000 output tokens. Knowledge cutoff: January 2026. Pricing: $10 / $50 per million tokens. No extra charge for longer context usage.
Great content! Keep up the good work!