Abhishek Soni

Claude Fable 5 : Anthropic Just Handed the Public a Restricted-Class AI

Abhishek Soni — Wed, 10 Jun 2026 04:32:22 GMT

On June 9, Anthropic released Claude Fable 5. It is the first “Mythos-class” model ever opened to the public. The same model class that autonomously found over 10,000 zero-day vulnerabilities in the world’s most critical software is now available to anyone with a paid subscription, wrapped in an AI firewall.

This is not another model launch. It is a category change in how we work with AI: from prompting to delegating.

1. AI graduated from prompts to projects

Every public model before this one worked in turns. You prompt, it responds, you correct.

Fable 5 runs inside an agent harness like Claude Code for days at a time. It plans across stages, spawns sub-agents, writes its own tests, and uses vision to compare output against the goal. Unsupervised.

The receipts so far:

Early testers report production bug backlogs cleared and full 3D projects built in one shot.
One CTO: apps that took roughly 100 prompts a year ago now get one-shotted.
On a frontier physics task, it reached in 36 hours roughly where GPT-5.5 landed after 4 days. On a third of the reasoning tokens.

Operator takeaway: stop scoping prompts and tasks. Start scoping delegable projects with a clear definition of done. That is the unit of work this model is built for.

2. The benchmark gaps are 2 to 5x, not 2 to 5%

The numbers that matter:

SWE-Bench Pro: 80.3% vs Opus 4.8’s 69.2%
FrontierCode Diamond (hard, maintainable agentic coding): 29.3% vs Opus 13.4% and GPT-5.5’s 5.7%. A 5x lead over OpenAI’s flagship.
First model past 90% on long-running analytics. A 10-point jump.
Every’s internal senior-engineer benchmark: 91/100 vs Opus’s 63.

Ethan Mollick says it outperformed every public model he has used “by a considerable margin.”

One honest caveat from CodeRabbit’s hands-on review: Fable 5 is brilliant at vague, exploratory builds, but for strict production code review, Opus 4.8 still feels safer. The lead grows with task length and ambiguity.

3. It beat Pokémon from raw screenshots. No tools. No maps.

Earlier Claude models needed a complex helper harness just to play Pokémon FireRed. Fable 5 finished the entire game on vision alone.

A silly demo with a serious point: it can also rebuild a web app’s source code from screenshots, and it reads charts and tables nested inside PDFs. If your work lives in documents (finance, legal, analytics, architecture), this is the biggest unlock per dollar.

4. The twin you can’t have

Here is the part most coverage buries.

Fable 5 and Mythos 5 are the same model. Mythos 5 simply has fewer restrictions, and what its preview version did explains all the caution.

Under Project Glasswing, a roughly $100M defensive program with partners like Microsoft, Google, Apple, Cloudflare and the Linux Foundation, the Mythos lineage autonomously found 10,000+ high and critical zero-days across major operating systems and browsers.

Mozilla patched 271 Firefox vulnerabilities from a single evaluation pass and shipped them in Firefox 150. That was 10x what Opus found. Where Opus built working exploits 2 times in hundreds of attempts, Mythos did it 181 times. Google Project Zero’s elite team averages about 170 finds a year. Mythos matched a decade of that output in weeks.

You are getting the defanged version of a near cyber-weapon. That is the real headline, and the reason the firewall, the 30-day data retention rule, and the price all exist.

The safety firewall: a separate AI decides which model answers you

Fable 5 ships with standalone classifier models screening every request in real time. Touch cybersecurity, biology, chemistry, or model distillation, and your query is silently answered by Opus 4.8 instead, with a notification.

An external bug bounty threw 1,000+ hours at it and found zero universal jailbreaks. External red teams also came up empty. Over 95% of sessions never touch the fallback.

The operator playbook

Match the model to the work, not the hype. Fable for ambitious, asynchronous projects it can break down, build, and verify alone. Opus for fast, synchronous collaboration. At $10/$50 per million tokens (2x Opus), using Fable 5 for chat is burning money.

Sell outcomes, not hours. Days-long autonomous agents change the agency business model. Scope a deliverable, deploy the agent, review the verified result. The margin lives in the gap between “AI-assisted” and “AI-delegated.”

Build cost controls before you scale. $50 per million output tokens on a model that runs for days compounds fast. Model routing, prompt caching, and hard spend caps are pre-launch requirements now.

Patch everything, then relax. The Glasswing disclosure wave means a flood of high-quality security patches is flowing into software you already run. Tighten patch SLAs for the next two quarters.

The bottom line

The teams that win the next 12 months will not be the ones writing better prompts. They will be the ones who learn to scope, hand off, and review multi-day AI projects like they would with a senior hire.

The question is no longer “what can AI answer for me?” It is “what can I fully delegate?”

What is the first project you would hand off to an AI for three days, no check-ins? Tell me in the comments. I read every one.

I break down what is actually shifting in AI, and how to build on it, every few days. No hype, just what operators need to know. If this was useful, subscribe and share it with one operator who needs it.

Abhishek Soni · LinkedIn · Instagram · Author of the upcoming book The Borderless Operator

The Metal Quietly Running the AI Boom

Abhishek Soni — Tue, 09 Jun 2026 14:00:38 GMT

And then a strange thought hit me, standing there holding it. This exact metal is inside almost every AI model I touch every single day. ChatGPT. Claude. Gemini. The tools I build client automations on. The tools running my second brain. All of them depend, in some quiet physical way, on the thing sitting in my palm.

Most people talking about AI in 2026 are arguing about which model is smartest. Almost nobody is talking about what these models are physically made of. That gap is exactly where the interesting story lives. So let me show you what I found.

Why a precious metal ended up at the center of artificial intelligence

Here’s the thing nobody tells you. AI does not run on software alone. It runs on data centers. Enormous, power-hungry, heat-belching buildings stuffed with chips that never stop working.

And those buildings have a favorite metal: silver.

Silver is the most electrically conductive element on earth. Roughly 7% better than copper, and the best heat conductor of any metal we know of. When you are pushing huge amounts of power through dense racks of chips that run hot all day, every connection point matters. Resistance builds heat. Heat kills hardware.

So silver shows up everywhere in the unglamorous plumbing: switchgear contacts, power distribution, busbars, connectors, the conductive pastes that bind chips to circuit boards, the thermal compounds that pull heat away from processors.

You can update software with a line of code. You cannot update silver. It has to be mined, refined, and delivered. That single fact changes everything.

Want a sense of scale? A single ChatGPT query uses close to 10 times the electricity of a standard Google search, according to Goldman Sachs research. Now multiply that across billions of queries a day. The bottleneck stops being intelligence. It becomes power, heat, and the metal that handles both.

The numbers that made me stop scrolling

I went down a research rabbit hole on this, and a few figures genuinely surprised me.

Silver demand for electronics and electrical use hit a record 680 million ounces in 2024, per the Silver Institute’s World Silver Survey. Every large data center now holds somewhere between tens of thousands and hundreds of thousands of ounces of silver across its systems. That’s not jewelry. That’s infrastructure.

And the trajectory is steep. Data center silver demand is growing 15 to 25% a year and is expected to roughly double by 2030 as the AI buildout continues.

Then there’s the price. In 2025, silver crossed $60 an ounce for the first time in history, more than doubling in a single year. It outpaced gold. It even beat a lot of the famous AI stocks everyone was posting screenshots of. The quiet trade beat the loud one.

This is bigger than silver. It’s a pattern.

Step back for a second, because the metal itself is not really the point. The pattern is.

Look at where the giants are putting their money in 2026. The big US tech companies are guiding toward somewhere around $1.43 Trillion(Gartner forecasts) billion in AI infrastructure spending this year alone. Roughly three-quarters of that goes to physical stuff, not models. Microsoft has openly said it has an $80 billion backlog of cloud orders it cannot fill, because there isn’t enough power.

Europe is pouring 20 billion euros into “AI gigafactories” and wants to triple its data center capacity. India has over $200 billion lined up across Google, Microsoft, Amazon, and Adani.

And here’s what almost every report quietly admits: all three regions are hitting the same wall. It isn’t chips. It’s power, grids, and physical materials. Capex announcements are easy. Transformers, switchgear, and conductive metal are not.

In every gold rush, the people selling shovels do just fine. But the smartest ones own the iron the shovels are made of.

Reality check: don’t fall in love with the story

Now let me be honest with you, because I’d rather you trust me than be impressed by me.

Silver is volatile. Really volatile. It hit a record near $120 an ounce in January 2026, then crashed about 30% in a single day. Bank of America even flagged it as one of the most bubble-like assets around at the peak.

So no, this is not me telling you to go buy silver. I’m not your financial advisor, and a video of a guy holding a bar is not a strategy.

The volatility actually teaches the more useful lesson. Silver’s market is tiny next to gold’s, so when leveraged traders rush in and out, the price whips three times harder in both directions. But underneath that noise, the slow fundamentals barely moved: a sixth straight year of supply deficit, and AI data centers pulling more of the metal in every quarter.

That’s the part founders should internalize. Don’t confuse volatility with a broken thesis. The price swings fast. The fundamentals move slow. You build on the slow thing.

What this actually means if you’re building something

I run an AI automation business. I spend my days on the very top layer of this stack, the software layer everyone can see. And this whole rabbit hole reminded me of something I keep relearning.

The obvious layer is crowded. Everyone is building chatbots and wrappers and “AI for X.” Margins there get thin fast, because there’s nothing scarce about it.

The durable advantage usually sits one layer down, in the thing everyone depends on and nobody is watching. For the AI economy, that’s power and conductive metal. For your business, it might be a distribution channel, a dataset, a relationship, a process nobody wants to do. The shape is the same.

So the question I’ve started asking myself, and the one I’ll leave you with: stop asking “which model is best.” Start asking “where is the scarcity?” That’s where pricing power lives.

The silver bar is back on my desk now. Still cold, still heavy, still a little scuffed. But I look at it differently. It’s a small reminder that the future may be digital, but it’s built on things you can hold in your hand.

Now I want to hear from you: What’s the most “invisible” thing your own business secretly depends on, the layer underneath that you rarely think about until it breaks? Drop it in the comments. I read every one.

Everyone Is Buying Tokens. Almost Nobody Is Shipping.

Abhishek Soni — Fri, 05 Jun 2026 07:19:32 GMT

There’s a number I can’t stop thinking about.

For every dollar a company spends on AI coding tools right now, roughly 18 cents turns into software that reaches a real user. The other 82 cents disappears into bug fixes, rewrites, reverts, and review cycles, much of it cleaning up code the same tools just generated. That figure comes from EntelligenceAI, which aggregated data across more than 2,000 companies using advanced AI coding tools.

Hold that next to the headline of the year: Anthropic is going public at a valuation approaching one trillion dollars, having blown past OpenAI to an annualized revenue run-rate around $45 billion, growing faster than any company in recorded history.

Both of these are true at the same time. The product is selling like nothing we have ever seen. And the people buying it, some of the most sophisticated engineering organizations on earth, are quietly admitting they can’t yet prove what they got for the money.

That contradiction isn’t a scandal. It’s a signal. And if you build, invest, or operate in this space, reading it correctly is worth more than any model benchmark.

The receipts

Let me put the evidence on the table, because the specifics matter more than the vibes.

Uber rolled out Claude Code to its engineers in December 2025. By March, adoption jumped from 32% to 84% of its roughly 5,000-person engineering org. Around 70% of committed code now originates with AI. Sounds like a triumph, until you learn the company burned through its entire 2026 AI coding budget in four months. “I’m back to the drawing board,” the CTO said, “because the budget I thought I would need is blown away already.”

Then came the part nobody at a vendor wants quoted. Speaking in late May, Uber’s COO Andrew Macdonald was asked whether all that spend was translating into better products for riders and drivers. His answer: “That link is not there yet.” The usage stats, he admitted, “make your head explode,” but he couldn’t draw a line from tokens consumed to features customers can feel.

Microsoft went further. It began revoking some engineers’ access to Claude Code, moving them onto its own cheaper Copilot CLI. The framing was “tool consolidation.” The subtext was the bill.

Meta built an internal leaderboard ranking 85,000+ employees by token consumption. Nvidia’s Jensen Huang has floated the idea of giving engineers token budgets as part of their compensation. Somewhere along the way, “how much AI did you use” quietly became a goal in itself, which is the exact moment a metric stops measuring anything real.

And the now-legendary data point: one developer running an autonomous agent framework racked up a $1.3 million monthly bill, 603 billion tokens across 7.6 million requests. The bill wasn’t the result of waste. It was the result of the tool working, running flat out, doing exactly what it was built to do.

This is what I’ve started calling the tokenmaxxing trap: when consumption becomes the scoreboard, you will always be able to consume more. The ceiling isn’t usefulness. It’s budget.

Why this keeps happening (it’s older than AI)

Here’s the part most hot takes miss. This is not an AI problem. It’s a productivity problem that economists have watched for a century.

There is a vast, treacherous gap between task-level productivity and economic productivity. AI can absolutely make the task of writing code faster. That is real and measurable. But shipping a product that customers value is not one task, it’s a chain: deciding what to build, building it, testing it, reviewing it, integrating it, maintaining it, and not breaking the eleven things connected to it.

Speed up one link in that chain and you don’t automatically speed up the chain. Sometimes you slow it down. One analysis of over 10,000 developers found that high-AI-adoption teams took on 47% more pull requests per day, and spent the gains on context-switching, orchestration, and reviewing machine-generated code. At the median organization studied, 44% of all engineering output was reactive, fixing or maintaining existing code rather than building anything new. As AI pushes more code out faster, reverts climb faster than output. Each revert spawns a bug-fix, each bug-fix adds to the reactive pile, and the loop compounds.

Steam power took decades to show up in productivity statistics, because factories had to be physically redesigned around it before the gains appeared. Electricity was the same. The technology arrives first; the reorganization that makes it pay off comes later, and it’s the hard part. We are living in exactly that lag right now, except it’s compressed into quarters instead of decades, which is why it feels like whiplash.

So is it a bubble? No, and that’s the wrong question

This is where I part ways with the doom crowd. The “82 cents is wasted” framing is catnip for people who want AI to fail. They’re misreading it.

When a genuinely general-purpose technology lands, electricity, the PC, the internet, a huge amount of early activity should look like waste. Thousands of people poke at it, try dumb things, build stuff nobody wants, and burn money learning what works. That experimentation isn’t the bug. It’s the price of discovery, and it’s how every general-purpose technology in history found its killer app.

The proof it can work already exists. Look at the one company that’s clearly winning: Anthropic shipped 120+ features in the first 90 days of 2026, more than one per working day, across Claude Code, its API, and its models. That’s the loop closing: more shipping creates more usage, more usage creates more feedback and revenue, which funds more shipping. The 18% problem isn’t that the tools don’t work. It’s that most organizations haven’t built the system that turns tool output into shipped value. Anthropic has. That’s the whole difference.

So the question isn’t “bubble or not.” The question is sharper and far more useful:

Where does the value actually land?

My real thesis: the value won’t come from “better Facebook”

Here’s the contrarian bet I’d put money on.

Most people assume that if AI makes software cheaper to build, we’ll get better versions of the software we already have, faster apps, slicker websites, smarter ad targeting. I think that’s mostly wrong, and it’s the most important thing to get right.

The incumbent software industry, the social networks, the e-commerce stacks, the SaaS giants, is in many ways a mature industry. Like steelmaking or the internal combustion engine, the core problems have been solved. Making those products 30% cheaper to maintain is nice, but it doesn’t create a trillion dollars of new value. It mostly compresses margins and quietly eliminates some jobs. That’s why Uber’s COO can’t find the link: he’s looking for AI to improve a product that is already near its local maximum.

The trillion-dollar value shows up somewhere else: in software that was never written because it was too expensive to justify.

Think about every piece of software that didn’t get built because the addressable market was too small to pay a team of engineers. The custom tool for a 40-person logistics firm. The niche internal system for a single hospital department. The hyper-specific app for a community of 10,000 people. The personalized software that serves exactly one company’s weird workflow. For the entire history of the industry, this “long tail” of software was economically impossible, the engineering cost exceeded the value created.

AI doesn’t make Instagram better. AI makes the impossible software possible. When the cost of building drops by an order of magnitude, an enormous category of software that was previously underwater suddenly floats. That’s not a feature on an existing app. That’s a new industry, and it’s where founders, not incumbents, will win.

This is why I’m not worried about the 18% number, and why I’m long on this technology while being short on the hype around the current use of it. The waste is real. The reckoning on token spend is real and overdue. But the discovery happening underneath the waste is the most exciting thing to happen to builders in twenty years.

What to actually do about it (operator’s checklist)

If you’re building or running a company right now, here’s how I’d translate all of this into decisions:

1. Kill consumption as a metric. Today. The moment “tokens used” appears on a dashboard as a goal, you’ve created a leaderboard for waste. Measure cost-adjusted outcomes instead: cost per shipped feature, cost per resolved ticket, cost per completed workflow, human time saved after review and correction.

2. Put a hard cap before you scale, not after. Uber found its budget gone in four months. A runaway agent loop with no ceiling is a seven-figure invoice waiting to happen. Set spend limits at 120% of expected budget on day one.

3. Route, cache, and batch ruthlessly. Use frontier models only where reasoning genuinely demands them; route everything else to cheaper tiers. Cache repeated context. Send anything latency-tolerant to batch inference at half the cost. For most agentic systems this cuts spend 40 to 70% with zero quality loss.

4. Build the loop, don’t just buy the tool. Anthropic’s edge isn’t access to a better model, it is the model, and it still wins on shipping discipline. The advantage is the system around the tool: clear specs, codebase standards, fast review, tight feedback from production. Buy the tool, but invest in the loop. The loop is the moat.

5. If you’re a founder, hunt the impossible software. Don’t build a thinner wrapper on an existing category. Find the software that couldn’t economically exist before this year. That’s where the next decade of value is hiding, and the incumbents structurally can’t chase it.

The one line to remember

AI didn’t make software free to build. It made a whole universe of previously impossible software suddenly cheap enough to attempt.

The companies measuring how many tokens they burned are asking the wrong question. The ones asking “what could we build now that we never could before?” are about to run the table.

The tokens were never the point. The systems are.

If this was useful, I write about AI, startups, and what’s actually working (vs. what’s just loud) for operators and founders. Follow along, and forward this to the one person on your team who keeps quoting the token leaderboard.

— Abhishek Soni