Anthropic Released Two Claude Models at Once: Why Only One of Them Is Open to You

Published on: 2026-11-16

Anthropic Released Two Claude Models at Once: Why Only One of Them Is Open to You

On 9 November, Anthropic shipped two Claude models in the same breath: Claude Fable 5.2 and Claude Mythos 5.2. Different names, different prices, different eligibility. Strip away the packaging and you find the same underlying model underneath.

That is the part worth sitting with. What changed on 9 November was not how clever the model got. What changed is how a model gets handed out.

One capability, two ways to receive it

📖 Glossary

AI Box (also known as Agent Computer / Agent PC), is a dedicated local hardware device that runs AI Agents. Pre-installed with an AI agent management system, plug-and-play, running 24/7. Users can remotely command AI to work via Discord, Slack, Telegram, WhatsApp, and more.

Claude Fable 5.2 is open to every paid plan and ships with built-in blocking for high-risk cybersecurity work and for bioscience research tasks. Claude Mythos 5.2 removes some of those restrictions, but it is not available on sign-up. You have to apply through Anthropic's Cyber or Life Sciences Verification Program and get approved before it opens up.

Pricing moved in your favour, at least on paper. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down 20% from the $10 and $50 of the previous release, and it handles a 1-million-token context window. It is already live on Claude Code, OpenRouter, Cursor, Perplexity, Google Cloud and OpenClaw.

In plain terms: one mind, two ways of issuing it. One version comes with a seatbelt and is for sale to anyone. The other has the seatbelt removed, and goes only to organisations that pass a credentials check.

The unlocked one is the one you cannot get

KAIHE AIBOX concept device in a living room beside a TWO KEYS layered flow panel

Here is the trap worth naming. If you only skim the headlines, the obvious conclusion is that the newer release is the stronger one and you should move to it. Put the two access conditions side by side and the picture inverts.

The version with the restrictions loosened is called Mythos. Mythos is also the version ordinary users cannot simply open. The one you can actually reach every day, Fable 5.2, is the one with the guardrails still in place.

This is not a company locking you out of AI. The two task categories being blocked are high-risk cyberattack work and bioscience research — exactly where vendors have historically been most cautious, and where a routine filter before access is unremarkable. But the structural shift is real: one model, several safety tiers, distributed according to use case and eligibility. The question for you stops being "which one is stronger" and becomes "which one was issued to me".

What the numbers do and do not settle

KAIHE AIBOX concept device on a night desk beside a MIXED SCORES holographic dashboard

Anthropic shipped a strong result sheet, but read the columns together and a few things need discounting.

First, topping charts and losing to someone can both be true. Fable 5.2 sits at the top of several leaderboards: 65 on the Artificial Analysis intelligence index, 1832 on the Arena WebDev coding board, 75 on AA-Robotics, where it also became the first model to score on the measurement-box task. It also completed 41.3% of real freelance projects on the Remote Labor Index, a benchmark built by Scale Labs and the Center for AI Safety — up from 2.5% on the same board a year earlier.

And yet on ARC-AGI-2, the widely respected AI reasoning benchmark, it scored 93.8% and placed third, behind GPT-6 Astra at 95.0% and GPT-6.1 Sol at 94.2%.

So "tops several leaderboards" is an accurate summary. "Best at everything" is an addition somebody else made, and the gap between those two sentences is precisely the tier you will actually work in.

Second, a price cut does not automatically mean a smaller bill. The input price fell 20%, but the same reviewers note the model is token-hungry: cost per task lands roughly where the previous version did. If you pay a flat subscription, more usage does not cost you more, and this is good news for you. If you buy API usage by the token, the discount is easily eaten by the extra tokens, and you may end up spending the same while feeling like you saved something.

Third, the most expensive tier should not be your default. Anthropic's own developer guidance recommends starting most coding and knowledge work on Opus 5.5 at medium effort, and Opus 5.5 costs half of Fable 5.2 per token. The official Claude Code walkthrough goes further, showing how to pick a tier and how to split one task across several models, with a self-test guide for your own workloads. A new model shipping is not an instruction to migrate tonight. One developer measuring the tiers also hit token exhaustion on the max setting, which is a fair reminder that the highest tier is not the most reliable one in every situation.

There is one more claim worth carrying, and it is not a benchmark. Anthropic says Fable 5.2 closed two open theoretical-physics problems connected to black holes, including a computation that had been unfinished for 23 years. Treat that as the company's own account until someone reproduces it.

Three questions for the next release

Whenever a new model ships, ask three things.

How does the price compare with the version you are on now? This time it fell 20%, and that is real.

Which tier should you actually start on? The official answer is not the priciest one. It is "begin in the middle, switch on task type", with instructions for how.

And for whom is the cheapness cheap? It holds for subscription payers. It may not hold for people paying per token. Know which one you are before you switch.

The single number I would remember is the 41.3%. It measures whether an AI can finish a piece of work a paying customer would accept — not whether it answers a multiple-choice question correctly. From 2.5% to 41.3% in a year is a more honest description of what changed than any percentage point on a synthetic leaderboard.

Fable 5.2 is available in OpenClaw. KAIHE AIBOX ships with OpenClaw preinstalled, so the machine is powered on, connected and running the new model — no environment to rebuild, nothing to migrate, nothing to reinstall. Models can live in the cloud while the machine that does the running stays yours.

Recommended Products

A1 Home Entry A1 Pro Enhanced A2 Professional A2 Pro Advanced X1 Enterprise G1 Flagship
© KAIHE AI - Agent Computer Specialist