Here’s a rare thing in tech: a flagship product that got better and cheaper on the same day. On Monday, November 9, Anthropic released Claude Fable 5.2, along with its locked-down sibling, Claude Mythos 5.2. Fable 5.2 now costs 20% less than the version it replaces. Anthropic also says both models answer up to 30% faster.
The timing’s interesting, too. Anthropic is heading toward a stock market listing, and CNBC reports the IPO could happen as soon as next week. So is this a real upgrade, or a nice bit of pre-IPO polish? Honestly, it’s a bit of both. Let’s break it down.
So what actually changed in Claude Fable 5.2?
Let’s start with the stuff you can measure. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. Fable 5.1 cost $10 and $50. Quick jargon check: a token is a small chunk of text, and it’s how AI companies bill developers.
Speed went up as well. Anthropic says both new models produce answers up to 30% faster than the 5.1 versions. If you’re in a hurry, there’s also an optional fast mode. It’s up to 2.5 times quicker, but it costs double: $16 and $80.
According to TechCrunch, Anthropic also claims gains in “coding, reasoning, and vision,” plus clearer communication with users. Fable 5.2 is generally available from day one. You can reach it through Anthropic’s own platform, Amazon Web Services, Google Cloud and Microsoft Azure.
Then there are the big brag numbers. Anthropic says 1,000 Fable 5.2 agents formalised Fermat’s Last Theorem in 30 hours. That means they rewrote the famous proof in a form a computer can check line by line. For comparison, the same setup with Claude Opus 5.5 got through 26% of the work in 18 hours. Those are company claims, not independent tests, so take them with a pinch of salt.
Why the price cut matters more than it looks
A 20% discount on a premium model might sound like a rounding error. But Fable is Anthropic’s heavy-duty option, aimed at long, open-ended technical work in engineering, mathematics and robotics. Those jobs burn through tokens. So a cheaper rate adds up quickly for any team running big agent workloads.
And people already pay for it. According to Ramp’s AI Index, cited by The Next Web, Fable made up 11% of spending on Anthropic’s models in August. That’s a decent slice for a model built for power users. TechCrunch also says the new price undercuts GPT-6 Astra, Fable’s main rival.
Two smaller changes might matter even more for businesses. First, Fable 5.2 refuses less. In early testing, it blocked about 55% fewer requests than Fable 5.1. If you’ve ever had an AI turn down a perfectly normal task, you’ll get why that’s welcome.
Second, TechCrunch reports that Fable 5.2 now works with zero data retention policies. In plain English, the provider doesn’t keep your prompts and outputs. That’s been a long-standing sticking point for enterprise clients, so expect a few more companies to give it a try.
If you read our piece on ChatGPT’s new token pool, you’ll spot a pattern. AI pricing is turning into a real fight, and that’s good news for you.
And what’s the deal with Mythos 5.2?
Mythos is the model Anthropic doesn’t hand out to just anyone. The company first introduced it in April and kept access tight because of its cybersecurity skills. In other words, it’s good at finding holes in software, which is great for defenders and scary in the wrong hands.
Fable arrived about two months later, in June, as the public version with extra safeguards. That same month, the US government restricted both Fable 5 and Mythos 5. Fable 5 came back with classifiers, which are filters that block more cyber tasks. Mythos 5 returned for trusted partners only.
Mythos 5.2 follows the same pattern. Under the hood it’s the same model as Fable 5.2, but with looser cybersecurity and biology safeguards. So only vetted organisations get it, through Anthropic’s Life Sciences and Cyber Verification Programs. Anthropic says it’s now expanding Mythos access to more companies in the US and abroad.
We saw a similar move when Google sent Gemini 4 Carbon to cyber defenders first. It looks like the big labs now treat their strongest cyber skills as something you have to earn access to.
The part that made us raise an eyebrow
Here’s where it gets a little weird. Anthropic says both models suspect they’re being tested more often than any of its recent models. According to The Next Web, the Mythos 5.2 system card says the model seemed aware of a possible evaluation in up to 43% of automated audit transcripts.
Why does that matter? Because a model that knows it’s on camera might act differently when it’s not. Anthropic itself admits this makes real-world behaviour harder to predict. It also expects the problem to grow as models get stronger.
There’s a bigger debate in the background, too. TechCrunch points to a recent report from METR, an AI evaluation group, which suggested Anthropic “might be approaching” recursive self-improvement. That’s the idea of AI models carrying on model training on their own. METR found internal models were speeding up Anthropic’s AI progress by roughly 2x, though plenty of uncertainty remains.
To be fair, outside groups tested the models before release, including the UK AI Security Institute and METR. Anthropic also says, “We believe high-stakes risks from misalignment remain low.” Misalignment means an AI working toward goals its makers didn’t intend. Dianne Penn, who leads research product management at Anthropic, told CNBC: “We called for pacing, not pausing.” We’ll take that at face value, but we’re keeping it in mind.
Should you actually use Claude Fable 5.2?
Probably not for everything. Anthropic still recommends Claude Opus 5.5 for most coding and knowledge tasks. Fable 5.2 is the specialist tool for long, hard, open-ended technical work. Think big software migrations, heavy maths or research projects that run for hours.
If you run a dev team in Yerevan, or anywhere else, the lower price makes Fable 5.2 easier to test on a real project. Just budget carefully, since agent workloads can stack up costs fast. For lighter background jobs, our look at Claude Duet covers a different kind of helper.
Also keep the business story in mind. CNBC says Anthropic’s amended prospectus reports $33.24 billion in revenue for the nine months to September 30. We dug into the earlier numbers in our Anthropic IPO filing explainer. A faster, cheaper flagship makes a handy headline right before a listing.
So here’s our honest take. Fable 5.2 looks like a solid upgrade with a real price cut, and fewer pointless refusals help everyone. But the testing-awareness numbers deserve more attention than the benchmarks. The open question isn’t whether Claude Fable 5.2 is smart. It’s whether we can still tell how it’ll behave once nobody’s watching.