Anthropic's Claude Fable 5.2 outputs up to 30% faster at 20% lower cost
Anthropic launched Claude Fable 5.2, generally available, and Claude Mythos 5.2, the same underlying model restricted to vetted cybersecurity and biomedical research organisations. Outputs come up to 30% faster than the 5.1 generation, and prices fell 20% — to $8 per million input tokens and $40 per million output tokens.
📊 Scorecard: our Bullish call on $AMZN is scored from $270.87, the last close before it ran. Its first result comes after the next close.
The News
Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 this week. Fable 5.2 is generally available — any developer can use it — while Mythos 5.2 is the same underlying model restricted to vetted cybersecurity and biomedical research organisations, meaning only approved groups in those fields can access it. One timing note: Anthropic's own announcements index dates the launch to November 9, with wide coverage following on November 10, so this is the cycle's biggest model launch surfacing this week rather than a single-day event. Anthropic is a private company, so the public read-through tickers here are Amazon, Alphabet and Microsoft — Fable 5.2 ships on Amazon Web Services, Google Cloud and Microsoft Azure, the three big cloud platforms.
The headline improvements are speed and price. Anthropic says the models generate outputs up to 30% faster than the 5.1 generation, and the price fell 20%: $8 per million input tokens and $40 per million output tokens, down from $10 and $50. A token is roughly a word fragment, the unit AI companies charge by. There is also an optional fast mode that runs at up to 2.5 times the speed for $16 per million input tokens and $80 per million output tokens — faster, but at double the standard price.
On Anthropic's own benchmarks, Fable 5.2 leads its rivals on the tests that matter most to developers. On FrontierSWE v2, a test of agentic coding — AI writing and fixing real software on its own — Fable 5.2 scored 69.9%, against 65.5% for OpenAI's GPT-6 Astra and 61.2% for Google's Gemini 4 Carbon. But the same table shows Gemini 4 Carbon winning on business workflows, scoring 53.9% on AutomationBench against Fable's 51.4%. Anthropic also says a team of 1,000 Fable 5.2 agents formalised Fermat's Last Theorem — a famous maths proof — in 30 hours, and that over 90% of merged changes to Anthropic's own core codebase are now written by Claude.
30% faster, 20% cheaper, and top of the coding benchmark — Fable 5.2 is Anthropic's strongest hand yet.
Why It Matters
This matters because coding is where AI models earn their keep. Developers pay for models that write working software faster and cheaper, and Fable 5.2 does both at once: up to 30% faster outputs and a 20% price cut. The FrontierSWE v2 lead — 69.9% against 65.5% for GPT-6 Astra and 61.2% for Gemini 4 Carbon — is the number developers will actually compare, because it measures the work they do every day. A model that wins the coding test and costs less is hard to ignore, even for teams already built around a rival.
It also matters because Anthropic is honest, in its own table, about where it loses. Gemini 4 Carbon beats Fable 5.2 on AutomationBench, the business-workflow test, 53.9% to 51.4%. That split — Anthropic wins coding, Google wins business workflows — is the real shape of the model race right now: no single model wins everything, and the winner depends on the job. The 1,000-agent Fermat proof and the claim that over 90% of Anthropic's own code changes are now written by Claude are the company's way of saying it trusts its own model with its hardest work.
For the public companies in this story, the read-through is about cloud, not Anthropic itself. Fable 5.2 ships on AWS, Google Cloud and Azure, so whichever model developers choose, one of those three clouds bills for the computing power behind it. Amazon closed Tuesday at $270.87, up 1.48%; Alphabet at $287.38, down 0.27%; Microsoft at $533.16, down 2.20%. The spread between them says the market is not yet picking a winner from this launch — the fight is between the model makers, and the clouds collect either way.
Analyst's Take
Our call is Bullish. Faster outputs, a 20% price cut, a clear lead on the coding benchmark developers care about most, and a 1,000-agent proof that the model can do sustained, coordinated work — Fable 5.2 is a serious release. The AutomationBench loss to Gemini 4 Carbon keeps it honest: Anthropic wins coding, Google wins business workflows, and developers will choose by the job.
What to watch: whether developers switch to Fable 5.2 for coding work at the new price, and whether the fast mode — up to 2.5 times the speed at double the price — finds buyers who value speed over cost. The model war is now a weekly event, and the next benchmark table is never far away.
Are you bullish or bearish on this story?
By The Numbers
FAQ
What is Claude Fable 5.2?
It is Anthropic's newest AI model, generally available to developers. It generates outputs up to 30% faster than the 5.1 generation and costs 20% less. Claude Mythos 5.2 is the same underlying model, restricted to vetted cybersecurity and biomedical research organisations.
What does 'up to 30% faster' mean?
Anthropic says the model produces its answers up to 30% more quickly than the previous 5.1 generation. There is also an optional fast mode at up to 2.5 times the speed, for double the price.
How much does it cost?
The standard price is $8 per million input tokens and $40 per million output tokens — a token is roughly a word fragment, the unit AI companies charge by — down from $10 and $50. Fast mode costs $16 and $80.
Does Fable 5.2 beat its rivals on every test?
No. On Anthropic's own benchmarks it leads on agentic coding, scoring 69.9% on FrontierSWE v2 against 65.5% for GPT-6 Astra and 61.2% for Gemini 4 Carbon — but Gemini 4 Carbon wins on business workflows, 53.9% to 51.4% on AutomationBench.
Why are Amazon, Alphabet and Microsoft the tickers?
Anthropic is a private company with no public stock. Fable 5.2 ships on Amazon Web Services, Google Cloud and Microsoft Azure, so those three public cloud companies are the read-through tickers — the public companies closest to the story.
👁
Source: Anthropic