Quick answer
Claude Fable 5.2 is, on current benchmarks, the strongest AI model you can actually use, but most people do not need it for everyday work. In his Fable 5.2 review, Matt Wolfe shows it topping most of Anthropic’s published benchmarks and the Artificial Analysis intelligence index. The biggest gains are on very hard problems such as scientific research, hardware design and complex builds. It costs $8 per million input tokens and $40 per million output tokens, double Claude Opus 5.5, and Anthropic itself recommends starting with Opus 5.5 for most coding and knowledge work. Use Fable 5.2 for your toughest, most open-ended jobs, or as an “advisor” that plans while cheaper models do the routine work.
Key takeaways
- Fable 5.2 and Mythos 5.2 launched on 9 November 2026. They are essentially the same model; Fable is the public version with extra cybersecurity guardrails.
- It leads most benchmarks, with the clearest jump in agentic scientific research (74.6%, against 58.7% for Opus 5.5 on the chart Matt shows).
- Everyday gains are modest. On coding benchmarks the improvement is a few points, and Opus 5.5 edges it on one knowledge-work test.
- Max effort is not always best. On one coding benchmark, “extra high” beat “max”, and Fable on “high” roughly matched Opus 5.5 on “max” for less money.
- It is expensive: about $26.54 per task at max effort on Artificial Analysis, the highest of any model listed.
- Paid Claude plans only. Free users cannot select it; launch week comes with extra usage.
In this guide
- Watch: the source video
- What is Claude Fable 5.2?
- Is Fable 5.2 really the world’s best AI model?
- How does Fable 5.2 score on benchmarks?
- Does more thinking make Fable 5.2 better?
- How much does Claude Fable 5.2 cost?
- Who can use Fable 5.2?
- What does Anthropic say Fable 5.2 can do?
- How did Fable 5.2 do in Matt Wolfe’s own tests?
- What have early testers built with it?
- Should you use Fable 5.2 or stick with Opus 5.5?
- How to try Fable 5.2 this week without wasting it
- Frequently asked questions
- The bottom line
- Related posts
Every few months a new model is called “the best AI model in the world”. This time it is Anthropic’s turn. Matt Wolfe, who runs the Matt Wolfe YouTube channel and the Future Tools site, spent his first few hours with Claude Fable 5.2 and published a review the same day. He did not get early access, so his tests are a genuine first look rather than a partner demo.
This is our plain-English Claude Fable 5.2 review, based on what Matt showed and what Anthropic has published. We separate the evidence (benchmarks and his own tests) from the claims (Anthropic’s showcase examples), and finish with a simple way to decide whether Fable 5.2 is worth your money.
Watch: the source video
What this video covers: Matt Wolfe reviews Anthropic’s new Claude Fable 5.2: benchmarks, pricing, access, his own game and hardware-gadget tests, and the best early community demos.
Video: Matt Wolfe (@mreflow, Future Tools). No paid sponsor in this video. This summary and the practical advice are our own; benchmark scores, prices and demo details are as shown or quoted in the video and in Anthropic’s launch post.
What is Claude Fable 5.2?
Fable 5.2 is Anthropic’s new top-tier Claude model, released alongside Mythos 5.2. According to Matt, the two are essentially the same model. Mythos 5.2 is the unrestricted version, available mainly to cybersecurity testers. Fable 5.2 is the one the rest of us can use, with additional guardrails around cybersecurity.
Anthropic positions it for “the most ambitious, open-ended technical work”. Its own advice is surprisingly restrained: for most coding and knowledge tasks, start with Claude Opus 5.5, its “daily driver”, and switch up only when you need to. It also suggests two specific roles for Fable:
- Advisor: smaller Claude models (Haiku, Sonnet or Opus) consult Fable when they get stuck.
- Orchestrator: Fable plans the work and hands simpler tasks to smaller, cheaper models.
If you have followed the earlier rumours, this is the successor to the Fable 5.1 model and separate from the Mythos 5.5 and Fable 5.5 leaks we covered earlier this month. Treat anything about later versions as unconfirmed until Anthropic publishes it.
Is Fable 5.2 really the world’s best AI model?
On the measurements available today, it has the strongest case, but “best” depends on the job. Two sources back the claim:
- Anthropic’s launch benchmarks, where Fable 5.2 leads in most categories.
- Artificial Analysis, an independent site that combines many benchmarks into one intelligence score. Matt shows Fable 5.2 at the top, ahead of Google’s Gemini 4 Carbon, which has been announced but not yet released to everyone.
Matt’s own view is more nuanced. He calls it “a big leap for really hard problems, but maybe not as big of a leap for knowledge work and coding”. In his game test he preferred the visuals from Opus 5.5. On his light-hearted BuseyBench drawing test, Fable 5.2 came only sixth. So it is the best at the hardest things, not automatically the best for your thing.
💡 Tip: Benchmarks are a starting point, not a verdict. Our guide on choosing an AI tool with benchmarks vs real-world results shows how to run a quick test on your own tasks.
How does Fable 5.2 score on benchmarks?
It sets new highs in most categories, with modest gains in coding and big gains in research-style work. Here are the figures Matt reads out from Anthropic’s chart:
| Benchmark (as shown in the video) | Previous / rival score | Fable 5.2 |
|---|---|---|
| Frontier Code (agentic coding) | 45.5% (Opus 5) | 49.1% |
| Frontier SWE V2 | 62% | 69% |
| Agentic scientific research | 58.7% (Opus 5.5), 68.1% (GPT-6 Astra), 64.8% (Gemini Carbon) | 74.6% |
| Business workflows | Gemini Carbon about 2 points ahead | Second place |
| One knowledge-work benchmark | Opus 5.5 slightly ahead | Second place |
Matt also notes new top scores in cybersecurity, protein design, agentic search, computer use, robotics, multidisciplinary reasoning and chart reading. His honest caveat: he puts “less and less weight on benchmarks” over time, and so should we. They show the size of the leap, not how the model will feel on your work.
⚠️ Warning: These numbers come from Anthropic’s own launch materials, as read in the video. Vendor benchmarks are chosen by the vendor. Wait for independent tests before treating any single score as settled.
Does more thinking make Fable 5.2 better?
Only up to a point. One of the most useful details in the review is about reasoning effort. On the Frontier Code 2.0 chart, Fable 5.2 scored higher on “extra high” than on “max”. Max lets the model think for longer, yet it scored lower and cost more.
There is a second useful finding. Fable 5.2 on “high” scored about the same as Opus 5.5 on “max”, but cost less. So for some coding work, a stronger model at a lower effort level can beat a cheaper model pushed to its limit, on both quality and price.
The practical lesson: do not default to max. Start at medium or high and only step up if the answer is not good enough. Our guide to choosing an AI reasoning effort level explains how.
How much does Claude Fable 5.2 cost?
Through the API, Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. Matt compares it with other options:
| Model (API prices as quoted) | Input per million tokens | Output per million tokens |
|---|---|---|
| Claude Fable 5.2 | $8 | $40 |
| Claude Fable 5.1 (previous) | $10 | $50 |
| GPT-6 Astra (OpenAI) | $10 | $50 |
| Claude Opus 5.5 | About half of Fable 5.2 (Matt: Fable is “double the price”) | About half of Fable 5.2 |
So Fable 5.2 is cheaper than the model it replaces and slightly cheaper than GPT-6 Astra, but still twice the price of Opus 5.5. Real-world costs add up quickly. On Artificial Analysis, it was the most expensive model listed at about $26.54 per task at max effort. Matt’s single BuseyBench drawing took 19 minutes 50 seconds, used 122,248 tokens and cost him almost $5.
One small plus: Artificial Analysis showed it using slightly fewer output tokens per task than Sonnet 5.5, so it is not wasteful with words. Before switching anything important, work out the cost on a real task with our guide to the real cost of an AI task.
Not sure which Claude model to pick? Our guide Which Claude model should you use? matches Haiku, Sonnet, Opus and Fable to everyday jobs in a simple table.
Who can use Fable 5.2?
Anyone on a paid Claude plan, plus developers through the API. According to the video, Fable 5.2 is available on all platforms from launch day. If you are on a paid plan, it should appear in the normal model dropdown in Claude.
- Free plan: no access to Fable.
- Pro and higher plans: access included. Matt quotes Pro at $17 a month; check the current price in your region on Claude’s pricing page.
- Launch week bonus: Matt says Anthropic is offering extra Fable usage resets and a percentage off weekly usage during launch week, so testing costs less of your allowance right now.
💡 Tip: Fable uses your allowance faster than smaller models. If you hit limits, our guide on what to do when your AI usage limits are cut has simple ways to stretch them.
What does Anthropic say Fable 5.2 can do?
Anthropic’s showcase examples are impressive, but remember they are hand-picked. Matt walks through three from the launch post:
- A prototype operating system. Anthropic says one engineer gave Fable 5.2 an empty folder and the model coordinated more than 1,000 sub-agents over eight days, producing an operating system that boots on a laptop, with drivers, file storage, windows, a login screen and an installer.
- Protein structure work. Anthropic says it combined two open-source protein-folding models to find a new way to predict a whole capsid structure (the protein shell of a virus).
- Hardware engineering. From a one-paragraph prompt, it reportedly designed three working devices with their own circuit boards, 3D-printed cases and software: a home voice-assistant hub, a USB speakerphone and a handheld virtual pet.
🔎 Unconfirmed / rumour: These are Anthropic’s own claims and demos, not independent tests. They show what is possible with expert set-up and plenty of compute, not what you will get from a single prompt on a Pro plan.
How did Fable 5.2 do in Matt Wolfe’s own tests?
Mixed but encouraging: excellent mechanics and planning, less impressive visuals. Matt ran two tests of his own.
Test 1: a 3D game in the style of Megabonk
He used his usual prompt asking for a clone of a popular game. Fable 5.2 politely declined to copy the original’s name, characters or story and built an original game in the same genre instead, which it called “Shardstorm”. Using the Claude Code app on medium effort, it took about 16 minutes and worked first time.
Matt’s verdict: the mechanics were “perfect”, with sprinting, stamina, levelling up and double jumps all working, but he felt Opus 5.5 does better graphics and game design. The sound was basic. He also wondered whether the speed was helped by light launch-day traffic. If you want to try this kind of build yourself, see how to build an original game with an AI coding agent.
Test 2: a real hardware gadget
Because Anthropic highlighted hardware design, Matt asked it to design a handheld desk gadget with a pixel-art character that reacts to button presses. He wanted a spec, real off-the-shelf parts, firmware code, a 3D-printable case and build instructions.
It delivered a parts list built around a Raspberry Pi Pico H, a small display, buttons and wiring supplies, plus four STL files for 3D printing, Python code and a wiring guide. It refused to draw a real person’s likeness and used an original robot character instead. When Matt asked for a fuller assembly manual on high effort, it took about 15 minutes and produced an illustrated step-by-step guide. He plans to build it, so the real-world result is still to come.
Fancy building something similar? Our step-by-step guide How to use AI to plan a simple electronics gadget shows you how to get a parts list, code and a 3D-printable case, and what to check before you buy anything.
What have early testers built with it?
Mostly games, apps and code-made video, and some of it is striking. Matt highlights posts from people who had early access. He links them all in his video description:
- Matthew Berman built a third-person shooter in the style of Gears of War with very polished graphics.
- Alex, from Matthew Berman’s team, produced around 23 demos, including a Photoshop-style editor, a Premiere-style video editor and a voice changer.
- Peter Yang asked for a 20-second travel ad for a shrine in autumn, made entirely in code with music, art and animation.
- Matt Shumer built “Mirror Earth”, a voxel-style open world whose weather matches the real Earth. Matt found it big and fun but glitchy.
Early testers often have more time, better prompts and higher effort settings than a first-day user. Matt suspects his own game might have looked better on “high” or “extra high”. If you create video, the code-made animation examples build on the workflow in editing video with Claude.
Should you use Fable 5.2 or stick with Opus 5.5?
For most people, Opus 5.5 is still the sensible default; Fable 5.2 is the specialist you call in. That is Anthropic’s advice and Matt’s conclusion. Here is a simple way to decide:
| Your task | Better starting point | Why |
|---|---|---|
| Emails, summaries, research notes, everyday writing | Sonnet or Opus 5.5 | Fable’s gains are smallest here and it costs twice as much as Opus |
| Small coding projects and website tweaks | Opus 5.5 | Benchmark gains are only a few points |
| Big, multi-step builds (games, apps, tools) | Fable 5.2 on medium or high | Strong mechanics and planning, worth it for complex jobs |
| Bugs you have not been able to fix | Fable 5.2 | Matt singles out hard problems and stubborn bugs |
| Hardware, science or engineering problems | Fable 5.2 | Biggest benchmark leaps are here |
| Running a team of AI agents | Fable 5.2 as planner, cheaper models as workers | Anthropic’s suggested “orchestrator” role |
Our new guide Which Claude model should you use? goes deeper, with a quick test you can run in 30 minutes. If you already use several models, our guide to routing simple tasks to cheaper models shows how to keep costs down.
How to try Fable 5.2 this week without wasting it
Pick one hard task, test it fairly and decide with evidence. If you are on a paid plan, here is a sensible plan for launch week:
- Choose one or two genuinely hard tasks you have struggled with, such as a stubborn bug, a complex build or a tricky planning problem. Easy tasks will not show the difference.
- Run the same prompt in Opus 5.5 and Fable 5.2, both on medium or high effort, not max.
- Compare quality, time and usage. Note how much of your allowance each run used.
- Decide where Fable earns its place, and keep everything else on Opus, Sonnet or Haiku.
- Use the extra launch-week usage while it lasts. Matt points out it will go away after this week.
💡 Tip: Keep your test prompts in a simple document. When the next “best model” arrives, you can rerun the same tests in minutes rather than starting again.
Ready to put this into practice? Start with our guide to choosing the right Claude model, then try planning a simple gadget with AI if you want to see what these models can do in the real world.
🛠️ Practical guides
Ready to put this into practice? These step-by-step guides answer two common follow-up questions:
- Which Claude Model Should You Use? Haiku, Sonnet, Opus or Fable — A simple table and a 30-minute test to match each Claude model to your everyday tasks.
- How to Use AI to Plan a Simple Electronics Gadget: Parts List, Code and 3D-Printed Case — Get a parts list, firmware and a 3D-printable case from AI, and check it before you spend a penny.
Frequently asked questions
What is Claude Fable 5.2?
Fable 5.2 is Anthropic’s top-tier Claude model, released on 9 November 2026 alongside Mythos 5.2. It is designed for hard, open-ended technical work and is available on paid Claude plans and through the API.
What is the difference between Fable 5.2 and Mythos 5.2?
According to Matt Wolfe, they are essentially the same model. Mythos 5.2 is mainly available to cybersecurity testers, while Fable 5.2 is the public version with extra cybersecurity guardrails.
Is Fable 5.2 better than Opus 5.5?
On most benchmarks, yes, with the biggest gains on very hard problems. For everyday coding and knowledge work the difference is small, Opus 5.5 costs half as much, and Anthropic recommends starting with Opus 5.5.
How much does Fable 5.2 cost?
API pricing is $8 per million input tokens and $40 per million output tokens. In the Claude app it is included with paid plans, but it uses your allowance faster than smaller models.
Can I use Fable 5.2 on the free Claude plan?
No. According to the video, you need a paid plan such as Pro or Max to select Fable.
Should I always use max reasoning effort?
No. On one coding benchmark, extra high beat max, and max cost more. Start at medium or high and increase only if you need to.
Was Matt Wolfe’s video sponsored?
There was no paid sponsor. Matt notes that Anthropic did not give him early access, so his tests were done after launch.
The bottom line
Claude Fable 5.2 has a strong claim to being the most capable AI model you can use today, especially for hard technical work, complex builds and hardware design. But it is expensive, its everyday gains are modest, and Anthropic itself says most people should start with Opus 5.5. The smart move is to treat Fable as a specialist: test it on your hardest task this week, keep notes, and let cheaper models handle the rest. Start with our guide to choosing the right Claude model.
Related posts
- Claude Mythos 5.5 Leak: Fable 5.5 Demos, GPT-6.2 Astra and This Week’s AI News
- Low, High or Max: How to Choose an AI Reasoning Effort Level
- How to Choose an AI Tool for Your Business: Benchmarks vs Real-World Results

[…] stress-tested on this blog. If you want to know how good the model is in general, start with our Claude Fable 5.2 Review: Is It Really the World’s Best AI Model?. This article looks at something narrower and more practical: what happens when you build a game […]
[…] setting, and Anthropic recommends medium or high for most work. We have a full write-up in our Claude Fable 5.2 review, so we will not repeat it […]