Anthropic shipped Claude Fable 5 on June 9, 2026, and it is not an incremental Opus update — it is a new tier above Opus entirely. Fable 5 is Anthropic's most capable generally available model to date, and the launch came with an unusual twist: the same model also exists in an unrestricted form called Claude Mythos 5, reserved for a small group of vetted customers. If you build with AI, automate creative pipelines, or just want to know whether Claude Fable 5 is worth $50 per million output tokens, here is everything confirmed so far.
What Is Claude Fable 5?
Claude Fable 5 is Anthropic's new flagship language model, positioned above the Opus 4.x family. Anthropic describes it as delivering "Mythos-level capabilities" with strong safeguards designed to make frontier-class intelligence safe for broad use. According to Anthropic, the model is state-of-the-art on nearly all tested benchmarks, with standout performance in software engineering, knowledge work, and vision.
The model is available through the Claude API under the model ID claude-fable-5, with a 1 million token context window and up to 128K output tokens per response.
Three capabilities define the release:
- Long-running, asynchronous execution. Fable 5 sustains complex coding and knowledge-work tasks for extended periods without human intervention — the kind of overnight, multi-hour agent runs that earlier models could not hold together.
- Advanced vision. It reads diagrams, charts, and tables nested inside files and PDFs, which opens up document-heavy work in finance, legal, analytics, architecture, and gaming. In coding, it implements visual designs with high fidelity and uses vision to critique its own output against the goal.
- Proactive self-verification. The model updates its own skills based on what it learns mid-task, and develops its own harnesses and evaluations to check its work.
Claude Fable 5 Release Date and Rollout
Claude Fable 5 went live on June 9, 2026 across several surfaces at once:
| Surface | Status at launch |
|---|---|
| Claude API (first-party) | Available, model ID claude-fable-5 |
| Amazon Bedrock | Available in US East (N. Virginia) and Europe (Stockholm) |
| Claude Platform on AWS | Available in North America, South America, Europe, Asia Pacific |
| GitHub Copilot | Generally available |
| Cursor | Available |
On Bedrock, access is expanding gradually across AWS accounts, and there is one notable requirement: you must opt into data sharing before invoking the model. Anthropic requires 30-day retention of inputs and outputs on Mythos-class models so it can detect patterns of misuse that are not visible from a single exchange.
Claude Fable 5 Pricing
On the first-party Claude API, Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens — double the rate of Claude Opus 4.8 ($5 input / $25 output). The 1M context window comes at standard pricing with no long-context premium.
There is a clever wrinkle in how safeguards interact with billing: when a harmful prompt gets routed to Opus 4.8 instead of Fable 5 (more on that below), you pay Opus prices for that response, not Fable prices. If a request is blocked mid-conversation, initial tokens bill at Fable rates and subsequent tokens at Opus rates.
For teams budgeting AI spend the same way you would budget credits on a creative platform, the practical takeaway is that Fable 5 is a premium tier you reserve for the work that justifies it — long agent runs, hard reasoning, vision-heavy document analysis — while cheaper models handle routine calls.
Claude Fable 5 vs Opus 4.8
Opus 4.8 was Anthropic's top model until last week, and it remains current — Fable 5 sits above it rather than replacing it. The differences that matter:
- Capability tier. Fable 5 is the new ceiling: state-of-the-art on nearly all tested benchmarks per Anthropic, with the biggest gains in long-horizon autonomous work and vision.
- Price. Fable 5 is 2× Opus 4.8 on both input and output tokens.
- API surface. The two are nearly identical to integrate. Both use adaptive thinking (
thinking: {type: "adaptive"}) and both reject the legacy sampling parameters (temperature,top_p,top_k). Fable 5 adds one quirk: explicitly sendingthinking: {type: "disabled"}returns an error — you omit the parameter instead. - Safety architecture. Opus 4.8 answers what it answers. Fable 5 ships with a fallback mechanism where certain categories of risky prompts are answered by Opus 4.8 instead.
If your workload already runs well on Opus 4.8, the upgrade case is long-running agents, document vision, and tasks where you were previously chaining multiple model calls to compensate for capability gaps.
Claude Fable 5 vs Claude Mythos 5: The Safeguards Story
The most interesting part of this launch is the two-model structure. Claude Fable 5 includes safeguards that limit its performance in specific areas where misuse risk is elevated: harmful prompts related to cybersecurity, biology, chemistry, and health fall back to a response from Opus 4.8 rather than Fable 5.
Claude Mythos 5 is the same model without those limits. It is available only to a small group of vetted customers — on Bedrock it sits in limited preview for cybersecurity and life-sciences use cases like vulnerability discovery, drug design, and biodefense screening.
This is a notable shift in how frontier labs ship: rather than holding the most capable model back entirely, Anthropic split the release into a broadly available safeguarded version and a tightly controlled unrestricted one. For everyone outside those vetted domains, Fable 5 is the frontier.
How Good Are the Benchmarks?
Anthropic's claim is that Claude Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional results in software engineering, knowledge work, and vision. Early independent impressions — from developer blogs to the Cursor and GitHub Copilot communities that got day-one access — center on the same theme: the model's edge shows up most in sustained, multi-step work rather than single-prompt benchmarks.
The vision claims are worth watching for anyone in creative or visual fields. A model that can look at its own rendered output, compare it against the design it was given, and iterate until they match is a meaningful change in how visual work gets reviewed — the same critique-and-refine loop that human art directors run on AI-generated image work today.
How to Access Claude Fable 5
Depending on where you build, you have four routes in:
- Claude API — use model ID
claude-fable-5with the Anthropic SDK. Adaptive thinking is the only thinking mode; omit thethinkingparameter entirely if you want it off. - Amazon Bedrock — model ID
anthropic.claude-fable-5(orglobal.anthropic.claude-fable-5via the Converse API), in US East and Stockholm regions. You must first opt into provider data sharing via the Data Retention API; there is no console UI for this setting at launch. - GitHub Copilot — generally available since June 9, 2026.
- Cursor — available in the model picker.
For AWS users without access yet, availability is expanding gradually based on Bedrock usage, and AWS Support can fast-track requests.
As model integrations like these land across the creative stack, we track the ones relevant to generation workflows in our product updates.
What Claude Fable 5 Means for AI Creators
Here is the part that matters if you spend your days generating rather than coding: Claude Fable 5 is a language model. It does not generate images or video. Models like Kling and GPT Image — the kind powering AI video generation and AI image generation workflows on platforms like Polyfaced — remain a separate category, and a new frontier LLM does not replace them.
What Fable 5 changes is everything around the generation step. Long-running autonomous execution means an agent can plan a content calendar, draft prompts, evaluate outputs, and iterate — for hours, unattended. Advanced vision means the model can actually look at a generated frame and judge whether it matches the brief, instead of guessing from a text description. Self-verification means fewer silent failures in automated pipelines.
The practical pattern we expect to emerge: frontier LLMs as the orchestration brain, specialized generation models as the hands. If you are already producing work in a creative studio workflow, the near-term win is smarter automation around the tools you use — not a new tool replacing them. Keep an eye on what creators are shipping as these orchestration patterns mature.
Frequently Asked Questions
Is Claude Fable 5 available now?
Yes. Claude Fable 5 launched June 9, 2026 on the Claude API, Amazon Bedrock (US East and Stockholm), Claude Platform on AWS, GitHub Copilot, and Cursor. Bedrock access is rolling out gradually across AWS accounts and requires opting into provider data sharing first.
How much does Claude Fable 5 cost?
On the first-party Claude API, Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens, with a 1M context window at standard pricing. When a prompt triggers the safety fallback to Opus 4.8, that response bills at Opus rates ($5/$25) instead.
What is the difference between Claude Fable 5 and Opus 4.8?
Fable 5 is a new tier above Opus 4.8 — state-of-the-art on nearly all tested benchmarks per Anthropic, strongest in long-running agentic work and vision, at twice the price. Opus 4.8 remains available and is also the model that answers when Fable 5's safeguards route a risky prompt away.
What is Claude Mythos 5?
Claude Mythos 5 is the unrestricted version of the same model, without the safeguards that limit Fable 5 in elevated-risk areas like cybersecurity and life sciences. It is available only to a small group of vetted customers, currently in limited preview on Amazon Bedrock.
Can Claude Fable 5 generate images or videos?
No. Claude Fable 5 is a text and vision language model — it can analyze and critique images, but it does not generate them. Image and video generation remain the domain of dedicated models like the ones behind Polyfaced's studio tools, which is why the emerging pattern pairs an LLM orchestrator with specialized generators.
The Bottom Line
Claude Fable 5 is the new frontier for generally available AI: a step up from Opus 4.8 in sustained autonomous work, document and design vision, and self-checking reliability — at a premium price and with a safety architecture that splits the truly unrestricted version off for vetted customers. Whether you write code, run research, or orchestrate creative pipelines, the model to benchmark your hardest workloads against just changed. And if your work ends in pixels rather than prose, the smart move is the hybrid one: let models like Claude Fable 5 do the planning and critique, and let purpose-built generators do the creating — starting with your next project.
Related reading: DiffusionGemma: Google's text-diffusion experiment · Ray 3.2: Luma's frame-directed video model
Sources: AWS News Blog — Claude Fable 5 on AWS, Claude Platform model documentation, GitHub Changelog — Fable 5 GA for Copilot.
