What is Claude Fable 5.2?
Claude Fable 5.2 is Anthropic's most capable generally available model, released on November 9, 2026. It replaces Claude Fable 5.1, which Anthropic now lists as a legacy model. Claude Mythos 5.2 is the same underlying model with looser cybersecurity and biology safeguards, open only to organisations vetted through Anthropic's verification programmes. Fable 5.2 reads text and images, writes text, and has a 1M-token context with up to 128K output tokens.
The price is $8 per million input tokens and $40 per million output tokens, down from $10/$50 on Fable 5.1, with cache reads at $0.20. Anthropic says it writes output up to 30% faster than Fable 5.1, and an optional fast mode raises output speed by up to 2.5 times at $16/$80. Thinking is always on; effort defaults to high on the API and medium in the Claude apps and Claude Code.
It is available through the Claude API, the Claude apps, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. It requires 30-day data retention, and Anthropic commits to keeping it available until at least November 9, 2027.
How much does Claude Fable 5.2 cost?
| Plan | Price | What you get |
|---|---|---|
| Input | $8 per 1M tokens | |
| Output | $40 per 1M tokens | |
| Cache writes | $10 (5 min) or $16 (1 hour) per 1M tokens | |
| Cache reads | $0.20 per 1M tokens | |
| Fast mode | $16 in / $80 out per 1M tokens |
API: $8 per 1M input tokens, $40 per 1M output tokens; cache writes $10 (5 min) or $16 (1 hour); cache reads $0.20; Batch API 50% off; fast mode $16/$80.
Compare this price with other models
Is Claude Fable 5.2 worth it?
Fable 5.2 fixes the problem that made Fable 5.1 hard to justify: it is now plainly better than Opus 5.5, and not only on Anthropic's say-so. Artificial Analysis ranks it first on its Intelligence Index at 65 on max effort, ahead of Gemini 4 Carbon at 60 and Opus 5.5 at 59. Anthropic's own table, which is self-reported, has it ahead of Opus 5.5 on FrontierSWE v2 (69.9% against 62.3%), Humanity's Last Exam with tools (74.2% against 67.7%) and the Remote Labor Index (41.3% against 21.3%). It still trails Opus 5.5 on the AA-Briefcase Elo rating, 1785 against 1800.
The per-token price is double Opus 5.5's, but the per-task price is the number that matters, and effort sets it. On Artificial Analysis's index, Fable 5.2 at medium effort scored 56 for $3.80 a task in 8.3 minutes, against $10.04 for Opus 5.5 at max effort and its 59. At max effort Fable costs $26.54 a task and takes 44.1 minutes, the most expensive entry on the index. Anthropic claims that at high effort it matches Opus 5.5's best FrontierCode 2.0 score (45.3% against 45.5%) for about half the cost per task; that is a vendor figure.
Against Fable 5.1 the changes are a 20% price cut, cache reads down from $0.25 to $0.20, output up to 30% faster and a new fast mode. The safeguards block the same things, but Anthropic counted about 55% fewer blocks in early testing across claude.ai, Claude Code and Cowork: about 60% fewer in cybersecurity and 40% fewer in biology. Blocked cyber requests are still answered by Opus 4.8 and research-biology requests by Opus 5. Anthropic also calls these its strongest protections against distillation, and one of them, preserved thinking, now applies to every account.
That anti-distillation change is the one engineers will notice. On Fable 5.2, a thinking block is only valid against the exact conversation that produced it, so an app that edits earlier turns, or changes its tool list between requests, gets a 400 error that retrying does not clear. The system card adds a quieter weakness: in prompt-injection scenarios Fable 5.2 falls back to Opus 4.8 more often than recent models do, which raised the attack success rate, even though it is among Anthropic's most robust models when it answers the attempt itself.
Where it earns its place
A head of engineering running Opus 5.5 at max effort for long coding agents
Get the same or better results without the max-effort bill
Test Fable 5.2 at medium or high effort on the same tasks. Artificial Analysis measured $3.80 a task for Fable at medium (score 56) against $10.04 for Opus 5.5 at max (59), and Anthropic claims high effort matches Opus's best FrontierCode score at about half the cost (vendor figure).
An ML platform team already paying for Claude Fable 5.1
Move to the newer model without breaking production
Change the model ID to claude-fable-5-2; at equal token counts a $20,000-a-month workload becomes $16,000. First run Anthropic's prefix check, because code that rewrites message history now gets 400 errors, and note that Fable 5.1 cannot read Fable 5.2's thinking blocks if a router falls back.
A computational materials-science group
Speed up simulation codes that eat cluster hours
Hand Fable the GPU optimisation work. Discovered Materials reports it made Quantum ESPRESSO's phonon code about 36 times faster, taking 13 benchmark cases from 3.3 hours to under four minutes; that is a partner claim in Anthropic's launch post.
A platform engineer whose agents run mostly on Sonnet 5.5
Raise quality on the hard steps without paying Fable rates on every call
Anthropic supports Fable 5.2 as an advisor that cheaper Haiku, Sonnet or Opus executors consult on harder problems. Sonnet 5.5 lists at $2/$10, so only the escalated calls pay $8/$40.
Before you commit
- Setup
- Model ID claude-fable-5-2 (anthropic.claude-fable-5-2 on Bedrock). From Fable 5.1 the migration is mostly drop-in, with four breaking changes: the preserved-thinking prefix check applies to every account, Fable 5.1 cannot read Fable 5.2's thinking blocks, the older computer-use tool gives way to the toolset except on Bedrock, and a Fable 5.2 executor accepts only a Fable 5.2 or Mythos 5.2 advisor.
- Integrations
- Claude API, Claude apps, Claude Code, Amazon Bedrock, Claude Platform on AWS, Google Cloud (Anthropic managed and Google managed) and Microsoft Foundry. Fast mode runs on Claude Code, the Claude Platform, both Google Cloud options and Anthropic-hosted Foundry. Forced tool use, prefill and temperature, top_p or top_k return a 400 error.
- Data and privacy
- Fable 5.2 is a Covered Model: prompts and completions are kept for at least 30 days for safety monitoring, and zero data retention is not available unless Anthropic expressly authorises it; a workspace without 30-day retention gets a 400 error. Enterprise Frontier Safeguards, in private beta, runs that monitoring on data stored in the customer's own infrastructure. Anthropic does not train on API data by default.
- Cost at scale
- An agent task with 500K input tokens (400K of them cached) and 50K output costs about $2.88 on Fable 5.2, against $3.60 on Fable 5.1 and $1.48 on Opus 5.5, before cache writes. A thousand such tasks a month is about $2,880; Batch halves it for work that can wait, and fast mode doubles it.
- Limits
- 1M context and 128K output. Rate limits run from 4,000 requests, 500,000 input and 100,000 output tokens a minute on the Start tier to 13,000 requests, 4M input and 800,000 output on Scale, in a bucket separate from Fable 5.1; Opus 5.5 gets 2M input tokens a minute on Start. A small set of AI-research topics, such as kernel work on some nonstandard accelerators, falls back to Opus 5. Retirement not before November 9, 2027.
What’s good
- First on Artificial Analysis's Intelligence Index (65 at max effort)
- 20% cheaper than Fable 5.1: $8/$40, with cache reads at $0.20
- About 55% fewer safeguard blocks than Fable 5.1 in Anthropic's early testing
- Optional fast mode with up to 2.5x output speed
What’s not
- Twice Opus 5.5's per-token price, and $26.54 per index task at max effort
- 30-day data retention required; no zero data retention without Anthropic's approval
- Cyber and research-biology tasks still handed to older Opus models
- Input rate limit a quarter of Opus 5.5's on the Start tier
Best for
Teams whose hardest coding, research or analysis agents still fall short on Opus 5.5, and who can accept 30-day data retention.
Claude Fable 5.2 alternatives
Claude Fable 5.2: common questions
- Is Claude Fable 5.2 worth it?
- The first Fable that independent testing puts clearly above Opus 5.5, and at medium effort it costs less per task than Opus does at max. We rate it 4.7 out of 5.
- How much does Claude Fable 5.2 cost?
- API: $8 per 1M input tokens, $40 per 1M output tokens; cache writes $10 (5 min) or $16 (1 hour); cache reads $0.20; Batch API 50% off; fast mode $16/$80.
- What are the downsides of Claude Fable 5.2?
- Twice Opus 5.5's per-token price, and $26.54 per index task at max effort; 30-day data retention required; no zero data retention without Anthropic's approval; cyber and research-biology tasks still handed to older Opus models; input rate limit a quarter of Opus 5.5's on the Start tier.
- Who is Claude Fable 5.2 best for?
- Teams whose hardest coding, research or analysis agents still fall short on Opus 5.5, and who can accept 30-day data retention.
- How much does Claude Fable 5.2 cost for a team or in production?
- An agent task with 500K input tokens (400K of them cached) and 50K output costs about $2.88 on Fable 5.2, against $3.60 on Fable 5.1 and $1.48 on Opus 5.5, before cache writes. A thousand such tasks a month is about $2,880; Batch halves it for work that can wait, and fast mode doubles it.
- Is Claude Fable 5.2 safe for company data?
- Fable 5.2 is a Covered Model: prompts and completions are kept for at least 30 days for safety monitoring, and zero data retention is not available unless Anthropic expressly authorises it; a workspace without 30-day retention gets a 400 error. Enterprise Frontier Safeguards, in private beta, runs that monitoring on data stored in the customer's own infrastructure. Anthropic does not train on API data by default.
- What are the best Claude Fable 5.2 alternatives?
- The closest alternatives we have reviewed are Claude Opus 5.5, Claude Fable 5.1, GPT-6 Astra and Gemini 4 Carbon.
Answers checked Nov 9, 2026.
Make Claude Fable 5.2? Show this rating on your site
