Claude Fable 5
Anthropic's most capable widely released model, released June 9, 2026 for the most demanding reasoning and long-horizon agentic work — thinking is always on, and single runs on hard tasks can execute autonomously for many minutes.
1M tokens · Text / Vision / Code · Prompt cache
Claude Fable 5 is Anthropic's most capable widely released model, sitting at the top of the lineup above Opus 4.8. It targets the hardest reasoning problems and long-horizon autonomous agentic work — complex refactors, overnight coding runs, and end-to-end deliverables that complete without human correction. It leads Anthropic's coding lineup on the harder SWE-bench Pro set and posts the company's strongest reasoning scores.
The defining trait is that thinking is always on — you don't configure it, and the raw chain of thought is never returned, only summaries. Fable 5 is priced at a premium above the Opus tier ($10 input / $50 output per 1M tokens), so the right call is to reserve it for the top of your difficulty range and route routine work to Sonnet 5 or Opus 4.8. On VM0 it carries the highest credit multiplier of any built-in model.
What is Claude Fable 5?
Released June 9, 2026 · Top of the Claude 5 family and of Anthropic's lineup — the premium frontier tier above Opus 4.8. Reserve it for the hardest reasoning and long-horizon agentic work.
Claude Fable 5 launched on June 9, 2026 as Anthropic's most capable widely released model. It targets the top of the difficulty range: multi-step reasoning, long-horizon autonomous agents, first-shot implementations of well-specified systems, and end-to-end enterprise deliverables (financial analysis, spreadsheets, slides, documents). Its stablemate claude-mythos-5 offers the same capabilities through Anthropic's Project Glasswing partner program.
The model had a turbulent early availability history. A US export-control order took Fable 5 and Mythos 5 offline on June 12, 2026; the order was lifted on June 30, and Anthropic restored Fable 5 on July 1, 2026 across Claude.ai, the Claude Platform, Claude Code, and VM0. It is fully available on VM0 today.
Two API behaviors set Fable 5 apart from the Opus tier. Thinking is always on — the thinking parameter is not configurable and an explicit disable is rejected — and the raw chain of thought is never returned, only a readable summary. Fable 5 also runs safety classifiers targeting research biology and most cybersecurity content, and requires 30-day data retention. It uses the same tokenizer as Opus 4.8, so token counts are roughly unchanged when moving between the two.
What's notable about Claude Fable 5
Headline architecture and capability features.
Fable 5 keeps the 1M-token context window (the maximum is also the default) and up to 128K output tokens. Thinking is always on and controlled through effort levels — low, medium, high, xhigh, and max — rather than a token budget. The raw chain of thought is never exposed; display: "summarized" returns a readable summary. Assistant-message prefill is not supported, and the model requires 30-day data retention (it is not available under zero data retention). Multimodal inputs span text, high-resolution vision, and code.
Specs at a glance
Claude Fable 5 benchmarks
Vendor-reported scores. Third-party trackers report different figures for the same benchmarks depending on dataset, context settings, and agent scaffold, so treat the headline numbers as configuration-dependent. Anthropic highlights the harder SWE-bench Pro set as the more durable coding signal.
Claude Fable 5 pricing
Provider list price, per 1M tokens.
How Claude Fable 5 behaves in practice
Observed behaviour from production agent runs.
Long-horizon autonomy
Fable 5 is state-of-the-art at long, autonomous agentic work — complex refactors and overnight coding runs that complete without human correction. Give it the full task specification up front in one well-specified turn and run at high or xhigh effort for the best results.
Always-on thinking
There is no thinking configuration to tune — the model always reasons before acting, and you control depth with the effort parameter. The raw chain of thought is never returned; you get a readable summary instead.
Parallel sub-agent delegation
Fable 5 reliably sustains ongoing communication with long-running sub-agents and peer agents. Asynchronous delegation — where sub-agents keep their context instead of re-establishing it per subtask — outperforms spawn-and-block patterns.
Safety classifiers and refusals
Fable 5 runs classifiers targeting research biology and most cybersecurity content, and can return a refusal on those requests. Benign adjacent work (security tooling, life-sciences tasks) can occasionally trip a false positive, so production pipelines should handle the refusal path and fall back to Opus 4.8 where appropriate.
Best agent tasks for Claude Fable 5
The overnight refactor that finishes on its own
Hand Fable 5 a large, well-specified migration and let it run autonomously across many minutes and hundreds of files. Its long-horizon coherence and leading SWE-bench Pro score are what pay off on runs that used to need a human babysitting each step.
The hardest reasoning problem you have
The teams with the best results gave Fable 5 their hardest unsolved problems first — let it scope the problem, ask clarifying questions, then execute. Its Humanity's Last Exam result is the strongest Anthropic has shipped.
The end-to-end enterprise deliverable
Financial analysis, spreadsheets, slide decks, and long documents produced in one run, with the model verifying its own output as it goes. Reserve it for deliverables where correctness matters more than token cost.
When to skip Claude Fable 5
Skip Fable 5 on routine, high-volume, or latency-sensitive work — its premium pricing ($10/$50) and always-on thinking make it expensive overkill where Sonnet 5 or Opus 4.8 hit the same bar for a fraction of the cost. Avoid it for research-biology or most cybersecurity workloads (its classifiers may refuse), and for organizations that require zero data retention, since Fable 5 mandates 30-day retention.
Claude Fable 5 vs other models
Claude Fable 5 vs Claude Opus 4.8
Fable 5 leads Opus 4.8 on the harder SWE-bench Pro set (roughly 80% vs 69%) and on reasoning, but costs 2× per token ($10/$50 vs $5/$25) and carries a ×4 VM0 multiplier vs Opus's ×2. Use Opus 4.8 as the default flagship for new agents; promote to Fable 5 only when a task is genuinely at the top of your difficulty range.
Claude Fable 5 vs Claude Sonnet 5
Sonnet 5 (×1) is the cost-efficient workhorse; Fable 5 is the premium ceiling. The common pattern is Fable 5 as the orchestrator on the hardest planning steps, delegating fan-out work to cheaper Sonnet 5 sub-agents. Reserve Fable 5's price for where it actually changes the outcome.
Claude Fable 5 vs GPT-5.5
Both are top-tier reasoning models. Fable 5 leads on Anthropic's highlighted SWE-bench Pro coding set and long-horizon autonomy; pick GPT-5.5 where its own agentic-terminal strengths dominate. On VM0 the choice is a routing decision, not a lock-in.
Claude Fable 5 vs Claude Opus 4.7
Opus 4.7 is a previous-generation flagship at Opus-tier pricing. Fable 5 supersedes it on the hardest reasoning and long-horizon work by a wide margin; keep 4.7 only for pinned agents you've already validated and don't want to re-baseline.
Bottom line: should you use Claude Fable 5?
The premium ceiling of the Claude 5 family. Reach for Fable 5 on the hardest reasoning and long-horizon agentic runs where correctness outweighs cost — and route everything else to Sonnet 5 or Opus 4.8. On VM0 it is the most expensive built-in model by credit multiplier, so use it deliberately.
Frequently asked questions
When was Claude Fable 5 released?
Anthropic released Fable 5 on June 9, 2026. It went offline under a US export-control order on June 12, was cleared on June 30, and was restored on July 1, 2026 across Claude products and VM0. It is fully available on VM0 today (model id claude-fable-5).
How does Fable 5 pricing compare to Opus 4.8?
Fable 5 is $10 per 1M input tokens and $50 per 1M output — roughly double Opus 4.8's $5/$25. On VM0 it carries a ×4 credit multiplier, the highest of any built-in model, versus Opus 4.8's ×2.
Can I turn thinking off on Fable 5?
No. Thinking is always on and the thinking parameter is not configurable — an explicit disable is rejected. You control reasoning depth with the effort parameter (low through max). The raw chain of thought is never returned; you receive a readable summary instead.
Why did a Fable 5 request get refused?
Fable 5 runs safety classifiers targeting research biology and most cybersecurity content, returning a refusal on those requests. Benign adjacent work can occasionally trip a false positive — production pipelines should detect the refusal and fall back to Opus 4.8.
Does Fable 5 work under zero data retention?
No. Fable 5 requires 30-day data retention and is not available under zero data retention. Requests from an organization that doesn't meet the requirement are rejected.
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 ×4 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 ×4 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 ×4, which means a step here costs 4× 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 July 1, 2026.