Claude Fable 5.2 launched on November 9, 2026 at $8 / $40 per million tokens with a 1M-token context window. Here is the spec sheet, Anthropic's benchmark table with its caveats, the first independent numbers, what is not published yet, and what it means for brands that want to be cited by Claude.
Client result: Wearview went from $0 to $36K MRR in 8 months.
This post is part of our AI SEO guide. Related reading: Claude Fable 5.2 pricing, ChatGPT Product Feed Requirements, AI Licensing Deals Boost ChatGPT Citations 48%, Only.
Abdelmoghit drives the content engine at AY Rank. He researches keywords, plans content clusters, and produces citation-optimized articles that rank in both Google and AI search engines.
Full Bio →Claude Fable 5.2 is Anthropic's model for demanding reasoning and long-horizon agentic work, released on November 9, 2026. It costs $8 per million input tokens and $40 per million output tokens, has a 1M-token context window, and uses the API ID claude-fable-5-2.
That is the short answer. This page is the long one: the spec sheet, the release timeline, Anthropic's benchmark table with the caveats that come with it, the first independent numbers, and a dated list of what nobody has published yet. Every benchmark, price and spec links to its source, and most of those sources are Anthropic's own pages.
The last section is for marketers. Claude Fable 5.2 can search the web and cite what it finds. If you want your brand in those answers, it helps to know what Anthropic documents about that process and where the documentation stops.
Facts checked 2026-11-09, between 21:00 and 21:30 UTC. Charts and Artificial Analysis numbers re-checked 2026-11-09 at 23:54 UTC. We will update this page as new numbers land.
Every row in this spec sheet comes from Anthropic's models overview, the Claude Fable 5.2 model page or the pricing page.
The same ID, claude-fable-5-2, works on the Claude API, Claude Platform on AWS, Microsoft Foundry, and both Google Cloud options. Only Bedrock uses a prefixed ID.
Two details on the model page matter for some buyers. Adaptive thinking "is always on and can't be turned off." You control depth with the effort setting. And Claude Fable 5.2 "requires 30-day data retention and isn't available under zero data retention unless expressly authorized by Anthropic," per the model page.
For the full cost breakdown with cache, batch and fast mode math, read our Claude Fable 5.2 pricing guide.
Anthropic released Claude Fable 5.2 on November 9, 2026. The @claudeai announcement post went out at 18:57 UTC, and the official announcement carries the same date.
It launched alongside Claude Mythos 5.2. Anthropic says the two "are the same underlying model." Fable 5.2 ships with cybersecurity and biology safeguards, so anyone can use it. Mythos 5.2 drops some of those safeguards and is limited to organizations accepted into Anthropic's Cyber Verification Program or Life Sciences Verification Program. Claude Enterprise customers can also reach it through Claude Security.
A warning on search results. When we checked Google at 21:29 UTC on launch day, the top results for "claude fable 5.2" still included a rumor page and videos made before Anthropic published any specs. The only launch date Anthropic has published is November 9, 2026.
The table below is Anthropic's own, copied from the announcement. Read the caveats under it before you quote a single cell.
Anthropic's table also has cybersecurity and protein design rows. Those report Mythos 5.2 scores, not Fable 5.2, so we left them out.
Fable 5.2 does not win every row. Gemini 4 Carbon leads AutomationBench, 53.9% to 51.4%. On the AA-Briefcase Elo line, Opus 5.5 sits at 1800 and Fable 5.2 at 1785.
The cost notes are more useful than the top scores. Anthropic writes that on FrontierCode 2.0, "At High effort, Fable 5.2 scores about the same as Opus 5.5's best score (45.3% vs. 45.5%) for roughly half the cost per task." On WANDR, Perplexity's benchmark for large data-collection tasks, Fable 5.2 at medium effort "matches Opus 5.5's top score for about 40% of the cost per task." So the model that costs twice as much per token can cost less per finished task. Our pricing guide, linked above, works through what that means for a real budget.
Artificial Analysis published its first Fable 5.2 numbers on launch day. When our team checked at 19:25 UTC, that page returned a 404. By 21:28 UTC it was live. We re-read it on 2026-11-09 at 23:54 UTC, and the charts and table below use those numbers.
Source: the Fable 5.2 model page on Artificial Analysis, checked 2026-11-09 at 23:54 UTC. These numbers move as Artificial Analysis reruns its tests. Between our first check at 21:28 UTC and this one, time per task went from 45.0 to 44.1 minutes and time to first answer token from 171.16 to 231.55 seconds. Artificial Analysis also calls the max-effort setup "notably slow per task and very verbose." It generated 275M output tokens across the index, against a median of 72.8M.
Read that as a trade. At max effort, Fable 5.2 tops this independent index, and it takes a long time and a lot of tokens to get there. At high effort it scores 62 for $9.08 per task. Opus 5.5 at max effort scores 59 for $10.04.
As of 2026-11-09, 21:30 UTC, we could not find a primary source for any of the items below. If you see a confident number for one of them, check where it came from.
Anthropic's own advice is to start with Opus 5.5. The models overview says: "If you're unsure which model to use, start with Claude Opus 5.5 for most workloads. Use Claude Fable 5.2 for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short."
The last row is easy to miss. Cache reads cost $0.20 per million tokens on both models, according to the pricing page. If most of your input is a cached prompt, the input-side gap between the two models shrinks.
Anthropic also describes two team roles for Fable 5.2. As an "Advisor," Haiku, Sonnet or Opus models consult it on hard problems. As an "Orchestrator," it hands simpler tasks down to smaller models.
Claude Fable 5.2 costs 2x Opus 5.5 per token. Anthropic's own data shows it can come within 0.2 points of Opus 5.5's best FrontierCode score at high effort for about half the cost per task. Test both on your own tasks before you switch.
When Claude Fable 5.2 runs with Anthropic's web search tool, it searches the web and cites the pages it uses. For a brand, the question is whether your page is one of the citations.
Below is what Anthropic documents about that process, all from the web search tool docs unless noted. None of it is specific to Fable 5.2. It applies to Claude models with the tool switched on.
The docs say Claude searches when a request depends on information that is "current, changing, or outside its training data." The examples Anthropic lists include recent news, current prices, and "information about specific organizations, people, or products that might have changed."
We read that list as covering most buying questions. "Which GEO agency should I hire" or "what does tool X cost" both depend on products and prices. Claude Fable 5.2's knowledge cutoff is June 2026, so anything your company shipped after that mostly reaches the answer through search.
Claude answers without searching for stable facts, math, coding concepts and creative writing. Developers can also nudge it to search more or less through the system prompt.
Citations are always on for web search. Each one carries the source URL, the page title, and cited_text, which is "up to 150 characters of the cited content." Anthropic also tells developers who show API output to end users that "citations must be included to the original source."
Our inference from that field limit, not Anthropic guidance: the sentence that answers the question should make sense on its own in about 150 characters. A claim that only works with the paragraph around it quotes badly.
With the web_search_20260209 tool version and later, "Claude can write and run code that filters the search results before they reach the context window." Anthropic calls this dynamic filtering. It is available on Claude 4.6 and later models, which includes Claude Fable 5.2, on the Claude API and most platforms.
So a page can come back in the raw results and still be filtered out before the model reads it. Anthropic does not document what the filter looks for, beyond keeping "only relevant information."
Developers building on Claude can limit sources with allowed_domains or blocked_domains, and localize results with user_location. A Claude-powered app in your market may only search a short list of domains. You can't see that list from outside.
Site owners control crawler access. Anthropic runs three bots, described in its crawler help article:
You can block ClaudeBot for training and still allow Claude-SearchBot and Claude-User. Check your robots.txt before you assume Claude can read your site. Our free robots.txt generator handles ClaudeBot and other AI training crawlers. Add the Claude-SearchBot and Claude-User lines by hand.
Anthropic does not say which search index the web search tool queries, how results are ranked, or what makes Claude prefer one source over another. Any article that lists "Claude ranking factors" as fact is guessing. We cover what the docs do say, in more depth, in our guide on how to get cited by Claude.
On one measure, yes. On WANDR, Perplexity's benchmark for large data-collection tasks, Anthropic reports 82.1% for Fable 5.2 against 72.3% for Opus 5.5 and 68.7% for Fable 5.1. The score says nothing about which websites it trusts or cites, so don't read a source-selection change into it.
robots.txt for Claude-SearchBot and Claude-User. Blocking them may reduce your visibility in Claude's search results, by Anthropic's own account.If you want a team to run this across Claude, ChatGPT, Perplexity and Gemini, that is what our GEO service does.
Claude Fable 5.2 is an Anthropic model for demanding reasoning and long-horizon agentic work, released on November 9, 2026. It costs $8 per million input tokens and $40 per million output tokens, has a 1M-token context window and 128K max output, and its API ID is claude-fable-5-2.
Claude Fable 5.2 was released on November 9, 2026. Anthropic's @claudeai launch post went out at 18:57 UTC that day, and the announcement, docs and pricing pages went live with it.
Claude Fable 5.2 costs $8 per million input tokens and $40 per million output tokens on the Claude API. That is 20% below Fable 5.1's $10 and $50. Fast mode costs $16 and $80, and the Batch API halves the standard price. Our pricing breakdown has worked examples.
Claude Fable 5.2 beats Opus 5.5 on most rows of Anthropic's benchmark table, but not all. Opus 5.5 is ahead on the AA-Briefcase Elo line, 1800 to 1785. Anthropic still recommends starting with Opus 5.5 for most workloads and moving up to Fable 5.2 when your evals fall short.
Not through the launch offers, which don't cover the Free plan. Anthropic's launch offer gives Pro plans and standard Team seats all Fable models until November 16, 2026, at 11:59 pm PT, capped at 50% of weekly usage limits. Max, Ultra and premium Team seats get one to three Fable limit resets instead.
They are the same underlying model with different safeguards. Fable 5.2 blocks high-risk cybersecurity and most research biology tasks and is open to everyone. Mythos 5.2 is for organizations accepted into Anthropic's Cyber or Life Sciences Verification Programs. Both cost $8 and $40 per million tokens.
Anthropic has not documented any change to how sources get picked for Fable 5.2. When Claude uses the web search tool, it decides when to search, can filter results with code on newer tool versions, and each citation carries the URL, title and up to 150 characters of cited text. Anthropic does not publish ranking factors. Our AI Visibility Report tests where Claude and other AI assistants cite your brand.
Facts checked 2026-11-09. Sources: Anthropic, Introducing Claude Fable 5.2 and Claude Mythos 5.2, Claude Platform models overview, Claude Fable 5.2 model page, Claude Platform pricing, Web search tool docs, Claude Fable 5.2 launch offers, Anthropic crawler help article, Artificial Analysis, Claude Fable 5.2, @claudeai launch post
| Item | Claude Fable 5.2 |
|---|---|
| API model ID | claude-fable-5-2 |
| Amazon Bedrock ID | anthropic.claude-fable-5-2 |
| Input price | $8 per million tokens |
| Output price | $40 per million tokens |
| Fast mode price | $16 input / $80 output per million tokens |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Thinking | Adaptive, always on, default effort high |
| Knowledge cutoff | June 2026 |
| Retirement | Not sooner than November 9, 2027 |
| Date | Event | Source |
|---|---|---|
| Nov 9, 2026, 18:57 UTC | Launch post from @claudeai | X |
| Nov 9, 2026 | Announcement, docs and pricing go live | Anthropic |
| Nov 9 to Nov 16, 2026 | Pro and standard Team seats get Fable models for one week, up to 50% of weekly limits | Claude support |
| Nov 9, 2027 | Earliest possible retirement date | Models overview |
| Benchmark | Fable 5.2 | Fable 5.1 | Opus 5.5 | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|---|---|---|
| FrontierCode 2.0 (agentic coding) | 49.1% (xhigh) | 32.4% (xhigh) | 45.5% | 42.8% | 41.1% (medium) |
| FrontierSWE v2 (agentic coding) | 69.9% | 56.3% | 62.3% | 65.5% | 61.2% |
| AutomationBench (business workflows) | 51.4% | 31.4% | 42.5% | 41.4% | 53.9% |
| WANDR, relaxed (agentic search) | 82.1% | 68.7% | 72.3% | not shown | not shown |
| Terminal-Bench-Science 0.1 (agentic science) | 74.6% | 52.6% | 58.7% | 68.1% | 64.8% |
| OSWorld 2.1, offline partial (computer use) | 89.3% | 83.5% | 87.2% | 79.5% | 81.5% |
| Remote Labor Index | 41.3% (high) | 17.9% | 21.3% | 20.8% | not shown |
| Humanity's Last Exam, with tools | 74.2% | 65.6% | 67.7% | 57.2% | not shown |
| RoboDojo Sim (robotics) | 64.9% | 27.6% | 39.4% | 29.8% | not shown |
| Chartography, no tools | 73.4% | 44.8% | 64.4% | 71.0% | 72.5% |
| AA-Briefcase v1.1, rubric | 66.1% | 61.5% | 61.2% | 52.0% | 61.8% |
| AA-Briefcase v1.1, Elo | 1785 | 1669 | 1800 | 1569 | 1611 |
| Artificial Analysis metric | Claude Fable 5.2 (max effort, default fallback) |
|---|---|
| Intelligence Index | 65, first of 74 models ranked |
| Fable 5.2 at xhigh and at high effort | 62 each, second and third |
| Next best non-Fable entries | Gemini 4 Carbon (high) 60, Claude Opus 5.5 (max) 59 |
| Output speed | 103 tokens per second |
| Cost to run one Intelligence Index task | $26.54 |
| Time per Intelligence Index task | 44.1 minutes |
| Time to first answer token | 231.55 seconds |
| Question | Opus 5.5 | Fable 5.2 |
|---|---|---|
| Price per million tokens | $4 / $20 | $8 / $40 |
| Default effort | medium | high |
| Fast mode price | $8 / $40 | $16 / $80 |
| Anthropic's positioning | "Strongest daily driver" | Most ambitious, open-ended technical work |
| Cache read price | $0.20 | $0.20 |
| Bot | What Anthropic says it does | Effect of blocking it, per Anthropic |
|---|---|---|
| ClaudeBot | Collects content that may be used for training | Signals that your future content should be excluded from training |
| Claude-SearchBot | Indexes content to improve search result quality | "May reduce your site's visibility and accuracy in user search results" |
| Claude-User | Fetches pages when a user asks Claude a question | "May reduce your site's visibility for user-directed web search" |