The Claude Mythos Moment

Anthropic’s Claude Mythos is the most capable AI model ever released. But the bigger story isn’t the model itself — it’s what it reveals about where the SaaS industry is headed. The cost of a unit of AI intelligence has fallen 150x in three years while capability has nearly tripled. For SaaS CEOs, this creates the most asymmetric opportunity since the shift to cloud: AI isn’t replacing software, it’s expanding what software can do. The winners will be the companies with distribution, brand, and fundamentals — not wrappers riding the hype.

---

This report is published by SaasRise, the #1 mastermind community for experienced SaaS CEOs with $1M–$100M+ in ARR. Members collectively have $3B+ in ARR.

---

The AI Era in Numbers

1. The Claude Mythos Moment — What Just Changed

On April 7, 2026, Anthropic first disclosed Claude Mythos — the most capable AI model ever built — initially making it available to select enterprise partners. On June 9, Anthropic publicly released Claude Mythos 5 alongside Claude Fable 5. At $10 per million input tokens and $50 per million output tokens, Mythos represents a step-change in what AI can do: deep multi-step research, agentic coding workflows, and reasoning at a level that consistently surpasses expert humans on the hardest benchmarks.

This isn’t just another model release. It’s a marker — the moment the SaaS industry has to reckon with a fundamental question: What happens when intelligence becomes a commodity?

To understand why that question matters, look at the progression. Claude has evolved from Haiku (a fast, cheap model for simple tasks) through Sonnet (the workhorse) and Opus (the thinker) to Mythos — a model that can read an entire codebase, understand the business logic, write tests, refactor the architecture, and deploy a working feature. Claude Code, Anthropic’s coding agent built on these models, hit $1 billion in annualized revenue within about six months of its May 2025 public launch — and by February 2026 had surged to $2.5 billion ARR.

For SaaS CEOs, the implication is immediate: your product roadmap just accelerated. What used to take a team of five engineers six months can now be scoped, prototyped, and shipped in weeks. The companies that recognize this — and embed AI deeply into their products and workflows — will pull ahead. The ones that treat it as a bolt-on feature will fall behind.

2. The Token Intelligence Curve — A New Law for the AI Era

There is a pattern emerging in AI that is as reliable as Moore’s Law was for semiconductors, and it may prove even more consequential. We call it the Token Intelligence Curve:

Consider: when GPT-4 launched in March 2023, it cost $60 per million output tokens. That was the only way to get GPT-4-level intelligence. Today, models that match or exceed GPT-4’s capability — like GPT-4o mini and GPT-5 nano — cost $0.40 per million tokens. That’s a 150x cost reduction in just over three years, while the frontier (Claude Mythos, GPT-5.2) has leapt far beyond what GPT-4 could do.

The numbers are striking. On SWE-bench Verified — a benchmark that measures an AI’s ability to resolve real GitHub issues — the best model scored 4.4% in mid-2023. By late 2024, that had jumped to 71.7%. As of June 2026, Claude Fable 5 leads at 95.0%, with Claude Opus 4.8 at 88.6% and GPT-5.5 at 82.6%. On GPQA, a graduate-level science benchmark, scores rose by nearly 49 percentage points in two years. Researchers have documented dramatic improvements in parameter efficiency — meaning models achieve the same capability with drastically fewer resources.

This isn’t just about price drops. It’s about the compounding of two exponential curves. When intelligence goes up and cost goes down simultaneously, the result is an explosion in what’s economically feasible. Use cases that were absurdly expensive 18 months ago — like having an AI review every customer support ticket and draft a personalized response — are now pennies per interaction.

3. AI Is a Subset of Software — The Optimistic Case

There’s a persistent narrative that AI will destroy the software industry. It won’t. AI is expanding it.

Technology analyst Benedict Evans, in a recent episode of Lenny’s Podcast (“The most rational take on AI you’ll hear this year,” May 2026), made this point with a vivid historical parallel. He showed an IBM ad from 1961 that reads: “An IBM electronic calculator… replaces 150 engineers.” The pitch is nearly identical to what you’d read in a Claude Code marketing page today. And yet, there are more engineers alive today than at any point in human history. The calculator didn’t eliminate engineering — it made engineering accessible to more people and expanded what engineering could accomplish.

“Even if models stopped getting better tomorrow, this is still an incredibly useful technology.”

— Benedict Evans, Lenny’s Podcast (May 2026)

Evans draws the VisiCalc analogy: when the first spreadsheet software arrived, people predicted it would eliminate accountants. Instead, it made financial modeling so cheap and fast that companies did more of it. The number of accountants grew. This is Jevons paradox applied to intelligence — when the cost of something drops dramatically, total consumption increases by more than the cost decrease.

This isn’t theoretical. We’re seeing it happen in real time. One SaasRise member — the CEO of a social media management SaaS company with over $20M in ARR — recently told the group: “We will not exist in five years as we exist today.” But rather than retreating, he’s investing. His team is rebuilding the entire product around AI, launching an AI-native version that does in seconds what previously required a marketing team and weeks of work.

As he put it: “I told the team we have to disrupt ourselves. That’s the first step.”

This is precisely the lesson from Clayton Christensen’s The Innovator’s Dilemma. I had Christensen as a professor at Harvard Business School in 2012 — about eight years before he passed away — and his core teaching was clear: the companies that survive disruption are the ones willing to cannibalize their own products before someone else does. The ones that protect their existing business at the expense of the future are the ones that die.

For SaaS CEOs, the optimistic case is this: AI isn’t shrinking the market for software. It’s making software able to do things that previously required expensive human judgment. Every workflow that was “too complex to automate” is now automatable. The total addressable market for software is growing, and the companies that embed AI into their products will capture a disproportionate share of that growth.

4. The New Moats — Distribution, Brand, and Data in the AI Era

If AI models are becoming commodities — and they are — then the competitive advantage shifts up the stack. The model is the infrastructure. The value is in what you build on top of it.

Benedict Evans frames this through the browser analogy: Microsoft won the browser war through distribution, bundling Internet Explorer with every copy of Windows. But the real value didn’t accrue to the browser maker — it went further up the stack to companies like Google, Facebook, and Amazon that built applications on top of the open web. The browser became invisible infrastructure.

We’re watching the same pattern play out with AI models. Google is spraying Gemini across every product. Meta is embedding Llama into Instagram, WhatsApp, and Facebook. Apple Intelligence runs on a billion edge devices. The model itself is becoming a utility — like electricity or cloud compute. Nobody asks which AWS region powers their favorite app. Soon, nobody will ask which LLM powers their favorite SaaS product.

For SaaS CEOs, this means the moat has shifted. The five defensible assets in the AI era are:

The model is commodity infrastructure — like AWS powering applications. The companies that own the customer relationship, the data, and the distribution will capture the lion’s share of value. The ones that are “basically reselling tokens,” as Evans puts it, will not.

5. Foundation Models Are NOT in a Bubble

Let’s address the elephant in the room. Are the big AI model companies overvalued? OpenAI at $852 billion. Anthropic at $965 billion after its May 2026 Series H. Are these bubble valuations?

No. And the revenue numbers tell you why.

Anthropic has executed what may be the most extraordinary revenue trajectory in the history of enterprise software. In January 2024, the company was doing roughly $87 million in annualized revenue. By December 2024, it hit $1 billion. By December 2025, it reached $9 billion. By April 2026 it crossed $30 billion, and as of its Series H announcement on May 28, Anthropic disclosed a run-rate of $47 billion in annualized revenue. OpenAI, meanwhile, has climbed to roughly $35 billion ARR as of May 2026. Combined, the two leading foundation model companies are generating over $82 billion in annualized revenue — a figure that was essentially zero three years ago.

Claude Code alone — a single product — hit $1 billion in ARR within roughly six months of its May 2025 public launch, then doubled to $2.5 billion ARR by February 2026. There is no precedent for this kind of revenue velocity in enterprise software. Slack took about seven years to reach $1 billion ARR. Zoom took about seven as well. Claude Code did it in six months — and then doubled it again three months later.

At its $965 billion valuation (Series H, May 2026) against $47 billion in ARR, Anthropic trades at roughly 20x revenue. That’s a premium — but for a company that grew revenue from $1 billion to $47 billion in 17 months, it’s remarkably restrained. For context, Salesforce trades at approximately 4x revenue with 12% growth. Anthropic is growing orders of magnitude faster, and the gap is narrowing fast: Anthropic has told investors it expects to be profitable in Q2 2026, with quarterly revenue projected at $10.9 billion — which would put it on a $43 billion+ annualized pace. By the time Anthropic IPOs, the multiple could compress to single digits.

OpenAI’s $852 billion valuation (its last private round; it has since filed a confidential S-1 targeting up to $1 trillion at IPO) against roughly $35 billion in current ARR is approximately 24x — pricing in the company’s massive consumer install base (ChatGPT with 900 million+ weekly active users), developer platform, and the prospect of achieving artificial general intelligence. OpenAI is projecting a $14 billion loss in 2026 as it invests in compute infrastructure, but revenue growth remains strong at $2 billion per month.

The fundamental thesis is straightforward: these companies are building the infrastructure layer that every application in the economy will eventually run on. If software ate the world, AI is eating software. OpenAI and Anthropic alone are already valued at nearly $2 trillion combined, generating over $82 billion in ARR. A combined valuation of $4 trillion or more for the leading foundation model companies is not unreasonable on a 3–5 year horizon if revenue trajectories hold — particularly with Anthropic achieving profitability and both companies approaching or exceeding $100 billion in annualized revenue by year-end.

6. The AI Wrapper Bubble — Where the Froth Is

If foundation models aren’t in a bubble, something else is: AI wrappers.

Across the venture capital landscape, a pattern has emerged. Startups that are essentially thin user interfaces over the ChatGPT or Claude API — with no proprietary data, no distribution moat, and no defensible technology — are raising $30 million rounds at $250 million pre-money valuations. Many are trading at 50–100x revenue multiples, if they have meaningful revenue at all.

These companies are, in Evans’s blunt framing, “low margin resellers.” When he was evaluating companies during the dot-com era, his test was simple: “You can say dotcom all you like — it’s a low margin reseller.” The same test applies today. You can say “AI-powered” all you like — if your core business is passing tokens from an API to a user interface and adding a markup, you are a low margin reseller.

The distinction that matters is between a wrapper and a harness:

• A wrapper is a thin UI over an API call. No proprietary data. No workflow integration. No switching costs. Minimal defensibility. Replace the API call with a cheaper model next quarter, and the product is instantly replicable.
• A harness is a product that uses AI as an ingredient within a deeply integrated workflow — with proprietary data, customer lock-in, regulatory compliance, API integrations, and domain expertise baked in. The AI makes the product better, but the AI alone isn’t the product.

Private equity firms are already pulling back. Several large PE shops have paused on AI-labeled deals because the unit economics don’t hold up under scrutiny. When you strip away the “AI premium,” many of these companies are 2x revenue businesses being priced at 15–20x.

This bubble will pop. It always does. And when it does, the companies with real revenue, real margins, and real moats will be the ones standing.

7. The Post-IPO Market Reset — Return to Fundamentals

We are about to witness three of the largest IPOs in history, all within approximately 12 months: SpaceX, Anthropic, and OpenAI.

SpaceX has already priced its IPO at $135 per share, valuing the company at $1.77 trillion — the largest IPO in history, raising $74.4 billion. OpenAI confidentially filed its S-1 on June 8, 2026, targeting a valuation of up to $1 trillion, with a listing expected in late 2026 or early 2027. Anthropic, now valued at $965 billion in private markets, is expected to follow with its own IPO in early-to-mid 2027, at a valuation that could exceed $1 trillion given its revenue trajectory.

These IPOs are historic events. They will also absorb an enormous amount of institutional capital — capital that is currently earmarked for “AI allocation” in venture portfolios, growth equity funds, and public market strategies.

Here’s the prediction: after these IPOs, the market will rotate back to fundamentals.

As one SaasRise member, a CEO in the hospitality tech space, noted: “The commentary from tech and SaaS stock analysts is increasingly a push back to fundamentals.” When institutional investors can buy liquid, publicly traded shares in the actual AI infrastructure companies, the incentive to pay 100x revenue for illiquid wrappers disappears.

The market will return to valuing what it has always valued in SaaS: revenue growth plus profitability. The Rule of 40 will reassert itself as the primary valuation framework. Median SaaS M&A will remain at approximately 4x annual revenue for companies with 30% growth and 10–15% net margins. Quality SaaS companies with strong fundamentals will always be worth 4x revenue or more. The AI hype premium will compress for everything except the actual foundation models.

8. The Benedict Evans Counterpoint — Will Model Companies Keep Their Margins?

Every honest analysis must steelman the opposition. Benedict Evans presents the most intellectually rigorous bear case for AI model companies, and it deserves serious consideration.

Evans’s thesis is simple: AI models are undifferentiated commodity infrastructure. As competition intensifies — with OpenAI, Anthropic, Google, Meta, Mistral, and dozens of open-source alternatives all converging on similar capabilities — pricing power evaporates. The models become interchangeable. And when that happens, value accrues up the stack, not to the model provider.

“Sam Altman says he wants to sell AI on a meter, like electricity. Let me explain to you the margin structure of the utility industry.”

— Benedict Evans, Lenny’s Podcast (May 2026)

Evans draws a devastating analogy with the telecom industry. Since 2010, mobile data consumption has grown by 1,500–2,000x. That’s extraordinary volume growth. And yet telecom stocks have gone nowhere in 25 years. The volume explosion was entirely absorbed by price competition, infrastructure costs, and capital expenditure. The money was real. The margins were not.

If you apply this logic to foundation models, the implication is sobering. OpenAI’s $852 billion–$1 trillion valuation implies massive future profitability. But if models become commodities — if the gap between the frontier model and the “good enough” open-source model continues to narrow (it’s already within 1.7% on key benchmarks) — then pricing power compresses, and the utility analogy holds.

The Counterargument

Here’s why I believe the telecom analogy, while instructive, is incomplete:

1. Network effects. Unlike telecom infrastructure, AI models benefit from data network effects. More usage generates more data, which generates better RLHF training, which generates better models. Anthropic’s Claude Code has accumulated billions of lines of real-world coding interactions that no competitor can replicate overnight.
2. Enterprise lock-in. Large enterprises don’t switch LLM providers casually. Fine-tuned models, custom system prompts, compliance certifications, and API integrations create meaningful switching costs.
3. Revenue velocity. Telecom companies grew revenue at single-digit percentages while capacity grew 1,000x. Anthropic grew revenue 30x in 15 months (from $1B to $30B). The growth rate itself is the moat — if you can grow into your valuation before margins compress, you win.
4. The “good enough” gap is not static. Every time the frontier models leap forward (as Mythos just did), the gap between “frontier” and “good enough” reopens. This is a treadmill that favors well-capitalized leaders.

The honest answer is that both can be true. Foundation models may compress on margins over the long term AND be worth trillions in the near-to-medium term, if volume grows fast enough to outpace margin compression. The key metric to watch is the spread between frontier and “good enough” model pricing. As long as frontier models command a meaningful premium — and Anthropic’s pricing for Mythos at $50/MTok vs. Haiku at $5/MTok suggests they do — the margin story holds.

9. How SaaS CEOs Should Position for the Mythos Era

The strategic implications of everything above crystallize into a clear playbook for SaaS CEOs:

Embed AI deeply — don’t bolt it on

The SaasRise member who said “we have to disrupt ourselves” has it exactly right. Adding an AI chatbot to your settings page isn’t a strategy. Rebuilding your core product experience around AI — so that the AI is the product, not an add-on — is. This is the Innovator’s Dilemma in real time: if you don’t disrupt your own product, someone else will.

Build all five moats

Distribution. Brand. Proprietary data. API integrations. Workflow embedding. Every quarter, audit your company against these five dimensions. If your only moat is “we use Claude/GPT better than competitors,” you don’t have a moat.

Use AI coding tools to respond at 10x speed

Claude Code, Cursor, and similar AI development tools are real. Engineering teams using them report 3–10x productivity improvements on certain categories of work. The SaaS companies that adopt these tools first will ship faster, respond to customer needs faster, and iterate on product-market fit faster than their competitors.

Stay obsessively focused on fundamentals

Revenue growth plus profitability. Rule of 40. Net revenue retention. Account churn. These metrics have always mattered, and they will matter even more as the AI hype premium fades. Quality SaaS companies with strong fundamentals will always be worth 4x+ annual revenue. That floor hasn’t changed, and it won’t.

Don’t panic about replacement

Evans makes a critical distinction: AI replaces tasks, not jobs. A single job is a bundle of dozens of tasks. AI can automate many of them, but the judgment, relationship management, and strategic thinking that tie those tasks together remain human. Your product isn’t being replaced by AI. Your product is being augmented by AI. The companies that understand this distinction will make better product decisions than the ones racing to “replace everything with AI.”

10. The 15-Year Outlook — Optimism Grounded in Reality

I want to end this report on a personal note. As a dad of a four-year-old and a one-year-old, I think about the long term differently than I did a decade ago. And I’m more optimistic about the next 15 years than I’ve ever been.

The rapid advancements in biology, medicine, software, and productivity — all of which are being accelerated by AI — point toward a world that is meaningfully better for the generation growing up right now. AI-assisted drug discovery is already cutting development timelines from a decade to years. AI-powered diagnostics are catching cancers that human radiologists miss. Software is becoming more capable, more accessible, and more affordable for businesses of every size.

The Token Intelligence Curve, projected forward, tells a remarkable story. If current trends hold — and there’s no physical law preventing them from holding for at least another decade — the cost of a unit of AI intelligence in 2036 will be approximately 600,000x cheaper than it was in 2023. Meanwhile, frontier capability will have increased by 1,000x or more. The intelligence available to a solo founder with a laptop in 2036 will exceed what the largest enterprises have access to today.

The Real Risks

Optimism doesn’t mean naivety. The risks are real, but they’re not the ones that dominate the headlines.

The existential risk narrative — AI achieving consciousness and deciding to eliminate humanity — receives outsized attention relative to its probability. The more pressing concern, raised by multiple SaasRise members, is the erosion of shared truth. As one member put it: “The erosion of democracy without a common truth” — the risk that deepfakes, personalized information bubbles, and AI-generated content make it impossible for citizens to agree on basic facts.

This is a solvable problem, but it requires intentional effort from technology leaders, policymakers, and companies like the ones in the SaasRise community. Evans’s advice is apt: “Presume radical uncertainty” about exactly how AI will reshape society, but recognize that historically, transformative technologies have created more value than they’ve destroyed.

11. Sources & Methodology