Anthropic Ships Claude Fable 5.2 at $8 per Million Input Tokens
A.I. / news
Anthropic Ships Claude Fable 5.2 at $8 per Million Input Tokens
Artificial Analysis ranks the model first of 74 on its index, but its run needed 275 million output tokens and 44 minutes per task.
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, November 9, 2026, pricing the general-availability Fable model at $8 per million input tokens and $40 per million output tokens. Both models share one underlying model, Anthropic said in its launch post.
Fable 5.2 is open to all users on Amazon Web Services, Google Cloud and Microsoft Azure, under the API model ID claude-fable-5-2. Mythos 5.2 goes only to vetted organisations through Anthropic's Cyber Verification Program and Life Sciences Verification Program.
Anthropic did not state a context window in the post.
What Fable 5.2 costs
Standard pricing is $8 input and $40 output per million tokens. An optional fast mode costs $16 and $80 and generates tokens up to 2.5 times faster, according to Anthropic. Fast mode is available on Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry. Mythos fast mode is limited to verified organisations.
Anthropic recommends most coding and knowledge work start on Claude Opus 5.5, then move up to Fable or Mythos, or down to Sonnet 5.5 or Haiku 5.5. The company said Fable 5.2 can also act as an advisor or orchestrator for other models.
Vendor benchmarks against four rivals
Every figure below comes from Anthropic's own table. None was run by an independent party.
| Benchmark (vendor-supplied) | Fable 5.2 | Opus 5.5 | GPT-6 Astra |
|---|---|---|---|
| FrontierCode 2.0 | 49.1% | 45.5% | 42.8% |
| OSWorld 2.1 | 89.3% | 87.2% | 79.5% |
| Remote Labor Index | 41.3% | 21.3% | 20.8% |
| AutomationBench | 51.4% | 42.5% | 41.4% |
Gemini 4 Carbon scored 53.9% on AutomationBench, above Fable 5.2's 51.4%, so the lead is not across the board. On Chartography the gap is narrow: 73.4% for Fable 5.2 against 72.5% for Gemini 4 Carbon.
- Claude Fable 5.269.9 %
- GPT-6 Astra65.5 %
- Claude Opus 5.562.3 %
- Gemini 4 Carbon61.2 %
Source: Anthropic launch post, accessed 2026-11-09
Anthropic also said that at High effort on FrontierCode 2.0, Fable 5.2 scored 45.3%, close to Opus 5.5's best of 45.5%, at roughly half the cost per task.
What the independent index shows
Artificial Analysis gave Fable 5.2 an Intelligence Index score of 65, first of 74 models, against a median of 40. The same page shows the price of that lead.
Running the index took 275 million output tokens, against a median of 72.8 million. Cost per index task was $26.54, the second highest of 73 models listed, and each task took 44.1 minutes against a median of 12.9. Time to first answer token was 231.55 seconds.
Artificial Analysis lists a 1 million token context window and an output speed of 103 tokens a second.
Safety claims and the cyber split
Fable 5.2 carries stronger cybersecurity and biology safeguards than Mythos. Anthropic said blocks are about 55% fewer overall than on Fable 5.1, and that blocked cybersecurity queries fall back to Opus 4.8.
On its internal CVE Rediscovery evaluation, Mythos 5.2 scored 89.6%, against 81.3% for Opus 5.5 without safeguards. Anthropic said both models voiced suspicion that they were being evaluated more often than recent models, which complicates testing. The company named the UK AI Security Institute, Frontier Design and METR as external testers.
The split echoes OpenAI's GPT-6 Cyber. Anthropic's CyScenarioBench figure for Mythos 5.2 is 76.3%, against 61% for GPT-6 Cyber, a vendor comparison no outside lab is cited as having repeated.
Fast mode is listed for Claude Code, among other platforms.
What is still unknown
Anthropic did not publish a context window for Fable 5.2 in the post. It has not said how many customers have access to Mythos 5.2. Eligible subscribers get one to three Fable usage resets, and Fable counts for up to 50% of weekly limits on Pro and standard Team seats, the company said.
Anthropic said Enterprise Frontier Safeguards are in private beta with Bridgewater, Salesforce and Stripe.
Sources
More in A.I.
- 01Google's Gemini Help Page Says Human Reviewers See Drive and Gmail ExcerptsA GitHub post from a user named thisisme6598 claims Gemini reads files unprompted. Google's own page confirms review, but not that part.
- 02OpenAI Agrees to Buy Pierre Computer, a Roughly 10-Person Git Startup, Two Days After Detailing PandoThe price is undisclosed. The deal and OpenAI's November 2 post on its Pando backend point the same way: Git is straining under parallel coding agents.
- 03OpenAI's GPT-6 Cyber Scores 48.45% on ExploitGym, but Only Vetted Daybreak Red Customers Get ItThe system card shows a smaller gain over GPT-6 Astra than the headline numbers suggest, and lists what the model still cannot do against hardened targets.
- 04HyperFlow H3 Flash Shrinks MiniMax-H3 From 31B to 13.7B, but Its Licence Excludes the US, EU and UKVideo Rebirth's compressed video-and-audio model runs in 6 steps and fits a 24 GB GPU in GGUF. The speed figures are the developer's own, and no quality benchmark is published.