Claude Fable 5
Anthropic's most capable publicly available model. Released June 9, 2026 as the first general release of the Mythos class, with state-of-the-art coding, knowledge work, and vision, and hard safety limits that fall back to Opus 4.8 on high-risk requests.
1M tokens · Text / Vision / Code · Prompt cache
Claude Fable 5 is Anthropic's most capable generally available model, released June 9, 2026 as the first public version of its restricted Mythos line. Anthropic describes it as state-of-the-art on nearly all tested benchmarks for software engineering, knowledge work, vision, and scientific research, with the lead over its other models growing as tasks get longer and more complex.
The headline trade-off is price and safety posture. Fable 5 lists at $10 per 1M input tokens and $50 per 1M output tokens, double Opus 4.8, and carries a x3.3 VM0 multiplier. In high-risk domains such as cybersecurity, biology, and chemistry it declines and falls back to Opus 4.8, returning a refusal rather than an error and not charging Fable rates for the rerouted request. Anthropic reports at least 95% of Fable sessions run entirely on the model's own responses.
What is Claude Fable 5?
June 9, 2026 · Anthropic's most capable publicly available model, positioned above the Opus 4 line. Ships on VM0 at a x3.3 multiplier; Opus 4.8 remains the recommended default for most agents.
Claude Fable 5 was released on June 9, 2026 as the first generally available model from Anthropic's Mythos class, which the company had previously held back as too powerful for unrestricted release. Fable 5 is the safety-gated public version; Mythos 5 remains restricted to approved organizations.
Anthropic positions Fable 5 as state-of-the-art on nearly every capability benchmark it tested, with the largest gains on long, complex, multi-step work. Third-party evaluators echoed this: Hex reported Fable as the first model to reach 90% on its core analytics benchmark of long-running analytical tasks, and Genspark said it beat every other model in its evaluations, with notably strong UI design and game-coding results.
The model ships with hard safety limits. In high-risk areas it blocks the response and falls back to Opus 4.8; Anthropic ran a 1,000+ hour external bug-bounty and additional red-teaming that surfaced no universal jailbreaks. Fable 5 and Mythos 5 are designated Covered Models with 30-day data retention and are not available under zero-data-retention terms.
What's notable about Claude Fable 5
Headline architecture and capability features.
Fable 5 keeps the 1M-token context window and up to 128K output tokens, billed at standard input pricing across the full window. It supports vision, tool use, memory, compaction, and adaptive thinking. High-risk requests are rerouted to Opus 4.8 and return stop_reason: "refusal" as a successful response rather than an API error, and rerouted requests are not billed at Fable rates.
Specs at a glance
Claude Fable 5 benchmarks
Vendor and third-party reported signals for Fable 5. Anthropic reports state-of-the-art results on nearly all tested benchmarks, with bigger leads on longer tasks. Treat these as directional and validate against your own VM0 agent regression set.
Claude Fable 5 pricing
Provider list price, per 1M tokens.
How Claude Fable 5 behaves in practice
Observed behaviour from production agent runs.
Frontier capability on long, complex work
Anthropic reports Fable's lead over its other models grows the longer and more involved the task. For multi-hour agent runs, deep research, and large analytical jobs, it is the strongest model Anthropic offers publicly.
Software engineering and analytics
Third-party evaluators put Fable at the top of their charts: first to 90% on Hex's long-running analytics benchmark, and the leading model in Genspark's agent evaluations, with standout UI design and game-coding results.
Vision and knowledge work
Native vision plus state-of-the-art knowledge-work scores make Fable a fit for workflows that mix documents, screenshots, and diagrams with high-stakes reasoning.
Safety fallback to Opus 4.8
In high-risk domains (cyber, bio, chem) Fable declines and reroutes to Opus 4.8, returning a refusal rather than an error. Anthropic reports at least 95% of sessions stay entirely on Fable, and you are not charged Fable rates for rerouted requests.
Cost posture
Fable lists at $10/$50 per 1M tokens, double Opus 4.8, and carries a x3.3 VM0 multiplier. It is a deliberate premium pick for the hardest work, not a default for routine agent loops.
Best agent tasks for Claude Fable 5
The multi-hour agent run that has to finish correctly
Long-horizon migrations, deep research, and multi-step analytical jobs where Fable's widening lead on complex tasks pays for the premium and a wrong answer is expensive.
The analytics workload at the edge of what models can do
Complex, long-running analysis of the kind Hex benchmarks. Fable was the first model to clear 90% there, so it is the pick when cheaper models stall on the hard cases.
The high-stakes UI or product-build task
Genspark singled out Fable for UI design and game coding. Reach for it when the build quality of the generated interface or app matters more than token cost.
When to skip Claude Fable 5
Skip Fable 5 for routine, high-volume agent loops where Opus 4.8 or Sonnet 4.6 already hit the quality bar at a fraction of the cost, for latency-sensitive chat where cheaper models are faster, and for cybersecurity, biology, or chemistry work that Fable will decline and reroute to Opus 4.8 anyway.
Claude Fable 5 vs other models
Claude Fable 5 vs Claude Opus 4.8
Opus 4.8 (x1.7) stays the recommended default for most agents and is what Fable falls back to on high-risk requests. Promote to Fable 5 (x3.3) for the hardest, longest tasks where its widening capability lead is worth double the vendor price; keep Opus 4.8 everywhere else.
Claude Fable 5 vs GPT-5.5
Both are top-tier frontier models. Fable leads on Anthropic's and several third parties' long-task and analytics benchmarks, but at a higher price (x3.3 vs x2). Pick Fable for the most complex multi-step work; GPT-5.5 remains strong and cheaper for agentic terminal coding.
Claude Fable 5 vs Claude Sonnet 4.6
Sonnet 4.6 (x1) is the workhorse default for most loops. The standard pattern is Fable 5 (or Opus 4.8) as the planner that delegates to Sonnet- or cost-saving workers, reserving Fable for the few steps that genuinely need frontier capability.
Bottom line: should you use Claude Fable 5?
The most capable model Anthropic offers publicly, and a deliberate premium pick. Reach for Fable 5 on the hardest, longest, highest-stakes work; keep Opus 4.8 as the default and Sonnet 4.6 as the cheaper workhorse beneath it.
Frequently asked questions
When was Claude Fable 5 released?
Anthropic released Fable 5 on June 9, 2026 as the first generally available model from its Mythos class. It is available on the Claude API (model id claude-fable-5), Amazon Bedrock, Vertex AI, Microsoft Foundry, and VM0.
How does Fable 5 pricing compare to Opus 4.8?
Fable 5 lists at $10 per 1M input tokens and $50 per 1M output tokens, double Opus 4.8's $5/$25. Cached input bills at $1.00 per 1M (a 10x discount). On VM0 it carries a x3.3 credit multiplier.
What happens on high-risk requests?
In high-risk domains such as cybersecurity, biology, and chemistry, Fable declines and reroutes to Opus 4.8. The API returns a refusal as a successful response, not an error, and you are not charged Fable rates for the rerouted request. Anthropic reports at least 95% of sessions run entirely on Fable.
How is Fable 5 different from Mythos 5?
Fable 5 is the safety-gated public release of the Mythos class. Mythos 5 is the same underlying technology without the public safeguards and remains restricted to organizations Anthropic has approved.
Should I use Fable 5 as my default model?
Usually no. Opus 4.8 is the recommended default for most agents; Fable 5 is a premium pick for the hardest, longest, highest-stakes tasks. Reserve it for steps that genuinely need frontier capability and let cheaper tiers handle the rest.
Does Fable 5 support prompt caching and vision?
Yes. It supports prompt caching (cached input at $1.00 per 1M, a 10x discount), a 1M-token context window, vision, tool use, and memory.
Alternatives
Using Claude Fable 5 on VM0
Two ways to access Claude Fable 5 on VM0
VM0 supports Claude Fable 5 as a Built-in model billed in VM0 credits, and through bring-your-own with a Anthropic API key. The Built-in path uses VM0 Managed routing and the credit multiplier explained below; the bring-your-own path bills you directly with the upstream vendor and skips the VM0 credit conversion entirely.
VM0's recommendation
VM0 positions Claude Fable 5 as a core agent model, recommended alongside Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 for the steps that drive the actual outcome of an agent run. These are the models we'd pick for the orchestrator role, for code-touching agents, and for any step where a wrong answer is expensive.
Credits and the ×3.3 multiplier
Every Built-in model on VM0 is priced as a multiple of Claude Sonnet 4.6, which sits at the ×1 credit baseline. Claude Fable 5 bills at ×3.3 credits. The multiplier is what shows up on your VM0 invoice; the vendor list price in the pricing table above is what the upstream provider charges before VM0 converts it into credits.
Claude Fable 5 bills at ×3.3, which means a step here costs 3.3× the credits of an equivalent step on Sonnet 4.6 (the ×1 baseline). It's a premium tier on VM0, so the cost-effective pattern is to default to a cheaper model and route only the steps that genuinely need the extra reasoning depth to Claude Fable 5.
Available on VM0 since June 9, 2026.