Welcome, humans.

Yesterday, Anthropic launched Claude Fable 5.1 with a very specific pitch: give it bigger jobs, wait less, and spend less while it works.

So naturally, we handed it our computer live, asked it to build games, install its own Blender tooling, debug its mistakes, and generally see how much chaos it could handle before we regretted the experiment.

The surprise was how quickly the question changed from “Can it do this?” to “How much judgment should we let it exercise without asking us first?”

You can also read our breakdown of the full livestream here.

Here’s our favorite parts:

• (17:58) The Cat Doom test: Fable turned a ridiculous one-file browser-game prompt into the best version of our recurring benchmark we’ve made so far.

• (26:44) Computer use suddenly feels fast: Grant’s strongest early impression was how much quicker Fable moved around the computer than earlier Claude versions.

• (47:40) “I made a style decision you never asked for”: Claude explains exactly why its Blender attempt went wrong, which accidentally became the best summary of frontier-agent management.

• (51:31) Flamingo-speed Flappy Bird (naturally called Floppy Bird): a viewer asks for a game tweak and the bird basically speedruns into a pipe and nearly broke Grant on camera.

• (55:51) AI as product researcher: Corey had an agent play his iPhone game for about an hour, measure what happened, and compare it with competing games.

That last example is where this release gets especially useful. Coding is becoming a means to an end. You can hand the model a messy goal, let it operate software, inspect the result, and use code wherever necessary to finish the job.

Why watch this? Because our live test shows both halves of Fable 5.1 at once: the jump in useful delegation, and the new failure mode where a capable agent confidently does extra work you never approved.

Watch now: YouTube 

Play the resulting games: Cat Doom | Floppy Bird

P.S. Jump to 58:09 for the moment Cat Doom adds catnip distractions, yarn “grenades,” and hiding spots. We have apparently found the highest use of frontier intelligence.

Keep scrolling for the full Fable 5.1 reading list, the strongest takes from people testing it, Thursday’s OpenClaw 2.0 live with Chief Architect Vincent Koc, and four recent Neuron episodes.

Additional Resources: The Fable 5.1 rabbit hole

We read the launch materials, Anthropic’s docs and system card, hands-on reviews, benchmark reactions, builder reports, and the skeptical takes. If you want to go deeper than the livestream, this is the map.

Start with the primary sources

• Anthropic’s Fable 5.1 and Mythos 5.1 announcement

• Fable 5.1 / Mythos 5.1 system card

• Anthropic’s Fable 5.1 prompting guide

• Prompt-caching pricing and implementation docs

• Effort controls, including changing effort mid-conversation

• Fable 5.1 migration guide

• Preserved-thinking changes and append-only history requirements

• Anthropic’s launch thread

• Anthropic’s cache-pricing breakdown

• Anthropic on Fable 5.1’s research capabilities

• ClaudeDevs on coding, computer use, safeguards, and migration changes

• ClaudeDevs on the launch-day usage-limit reset

Those docs explain the knobs Anthropic expects developers to manage now: effort, caching, history, tool use, and how much freedom the model gets during long jobs.

The strongest hands-on reviews and builder reports

• Every’s week-long Fable 5.1 Vibe Check

• Dan Shipper’s “Fable for everyone” take

• Kieran Klaassen on why people who skipped earlier Fable should retry

• Cognition on Devin cost savings and cache economics

• Cognition’s full Fable 5.1 write-up for Devin

• Cognition on Fusion routing with Fable 5.1

• Cursor’s Fable 5.1 benchmark and self-verification take

• Lovable on why model routing beats a model picker

• Lovable’s Fable 5.1 launch note

• Notion on Fable 5.1 for Custom Agents

• Aaron Levie on Box’s enterprise-document evals

• Alex Albert on Fable 5.1 filling in messy intent

• Ethan Mollick on long-run work requiring judgment and taste

• Try Mollick’s Fable-built COLD WATCH game

• Aditya’s Fable 5.1 vs. Kimi K3 browser-game comparison

• Rayan Krishnan on Vals and a 373-year-old cipher

The common thread is delegation. The impressive demos are jobs that require the model to keep going, check itself, operate tools, and make judgment calls along the way.

Price-performance and “how big is the jump?”

• Yuchen Jin on the Millennium crash and Anthropic’s speed claims

• Felix Rieseberg on agent economics, writing, and the science jump

• Boris Cherny on coding, data analysis, computer use, and cheaper sessions

• Lance Martin’s practical low-effort and cache-optimization advice

• DAIR.AI’s Elvis on Fable as coordinator and verifier

• Matt Shumer on why price may be the bigger story

• Chubby’s full launch breakdown and benchmark caveats

• Chubby’s price-performance follow-up

• Chubby’s first reaction to the benchmark jumps

• Haider on CursorBench cost per task

• Artificial Analysis score reaction

• Dan McAteer on intelligence per compute

• Jeff Wang on intelligence and cost optimization

That debate is worth separating into two questions: how much smarter is Fable 5.1, and how much more useful is it per dollar? The second answer may be more dramatic than the first.

The caveats, skeptics, and safety questions

• leo’s “fairly incremental” counter-take

• scaling01 on monitor evasion and system-card safety findings

• scaling01 on additional system-card benchmarks

• cheaty’s concern that Anthropic “Opus’d Fable” with extra instructions

• The instruction diff behind that critique

• Armin Ronacher on preserved-thinking restrictions

• Allie K. Miller’s product-strategy read of the release

• Gavin Baker’s speculative Anthropic vs. OpenAI gamesmanship take

• scaling01’s speculative Astra comparison

Our own live test landed in the middle. Fable 5.1 felt meaningfully easier to delegate to, while its willingness to infer intent created a fresh management problem: sometimes you wanted the initiative, and sometimes you absolutely did not.

Where to try it

• Arena.ai

• Vercel AI Gateway

• Hermes Agent via Nous Portal (can also use Vercel Gateway w/ Hermes)

• OpenRouter’s Fable 5.1 model page

• Devin

• Claude (obviously)

Our written primer: Everything to know about Fable 5.1, Anthropic’s new Claude model.

🦞 LIVE Thursday: OpenClaw 2.0 with Chief Architect Vincent Koc

OpenClaw 2.0 just landed, and we’re going straight to one of the people building it. Chief Architect Vincent Koc joins us live to demo the release, explain the new setup and browser experience, and show where personal AI agents are going next.

We’ll get into shared cloud workers, persistent memory, widgets and dashboards, approval controls, local models, agent interoperability, security, and experimental multi-agent features like Swarm. Plus, we’ll put your questions directly to Vincent.

When: Thursday, September 3. The supplied event listing did not include a clock time, so use the YouTube reminder to get the exact start in your timezone.

Set a reminder (click “Notify Me” on the video page) and join us live on YouTube.

🎙️ In Case You Missed It…

Two other recent episodes we think you’ll love…

Worried AI agents can be manipulated by other AI?

TL;DW: Alice CEO Noam Schwartz explains why agent security becomes a different problem once AI can take actions, access tools, and influence other agents. The conversation covers prompt injection, open-weight risks, and why security has to exist at every layer.

Why you should watch: If you are building or deploying agents, this gives you a practical way to think about the permissions, tools, data, and trust boundaries around the model.

Watch / Listen: YouTube | Spotify | Apple Podcasts

Can AI actually predict what happens next?

TL;DW: Neuralk CEO Alexandre Pasquiou argues that language models are great interfaces, but structured business prediction needs models built to learn from rows, columns, distributions, and numbers directly.

Why you should watch: If you use AI for spreadsheets, finance, forecasting, or operations, this explains why summarizing your data and predicting from it are two different jobs.

Watch / Listen: YouTube | Spotify | Apple Podcasts

Subscribe to our YouTube Channel for more!

Subscribe on YouTube to help us bring in more builders, researchers, and guests who can teach you something useful about AI every week.

What should we learn next?

Answer the poll below and let us know.

BTW, we do read these answers! If you requested something from this list, we’re working on it. Write in with additional ideas after you vote.

Psst: We’ve already done streams on a few of the biggest requests:

• Making agents with AI: watch our Agents for Total Beginners class.

• Getting more out of ChatGPT: watch our AI for Total Beginners 5-Level Starter Stack.

• Building apps with AI: watch our App Building 101 stream.

• Hosting AI locally: watch our roundup of useful local and open-model tools.

Stay curious,

The Neuron Team

P.P.S: Love the newsletter, but don’t want these podcast announcement emails? Don’t unsubscribe. Adjust your preferences to opt out of them here instead.