Does Anthropic make more money than OpenAI in 2026?

On reported annualized run rate, yes. Anthropic's run rate surpassed $65 billion at the end of July 2026, reported by Bloomberg on August 17, while OpenAI's latest reported run rate was around $40 billion in August 2026. Run rate annualizes a recent short period rather than measuring trailing revenue, so the figures are directional rather than audited, and neither company has published financial statements.

How fast is Anthropic growing?

Anthropic's reported run rate went from roughly $9 billion a year earlier to $47 billion in May 2026 and past $65 billion by the end of July 2026. That is roughly 38% growth in two months on an already very large base, which is the part investors find remarkable — high growth rates normally decay as the denominator grows.

What is Perplexity's revenue compared to the model labs?

Perplexity's annualized revenue passed $750 million in August 2026, up from under $250 million at the start of the year. That is roughly 1% of Anthropic's run rate and under 2% of OpenAI's, but Perplexity is an application company buying inference rather than a lab selling it, so the comparison measures different layers of the stack rather than direct competitors.

Is annualized run rate a reliable measure?

It is directional at best. Run rate takes a recent month or quarter and multiplies it out to a year, so it flatters any company with a steep growth curve or lumpy enterprise contracts, and it says nothing about profitability. All three figures here come from press reporting of private disclosures rather than audited statements. Use them to compare trajectory, not to value the companies.

Quick Answer

Anthropic vs OpenAI vs Perplexity Revenue, Aug 2026

Published: August 25, 2026

The Short Answer

As of August 2026, on reported annualized run rate:

Anthropic — $65B+ (end of July 2026, reported August 17)
OpenAI — ~$40B (August 2026)
Perplexity — $750M+ (August 2026)

Anthropic has, on this metric, overtaken OpenAI. All three numbers come from press reporting of private disclosures, not audited accounts.

The Numbers and Their Trajectories

Company	Run rate	Prior data point	Reported
Anthropic	$65B+	$47B (May 2026), ~$9B (mid-2025)	Bloomberg, Aug 17, 2026
OpenAI	~$40B	~$25B flat through spring 2026	Aug 2026 reporting
Perplexity	$750M+	<$250M (Jan 2026)	The Information, Aug 2026

The growth rates matter more than the levels:

Anthropic: ~38% in two months, on a base already above $47B
OpenAI: roughly 60% since spring, after a flat period near $25B
Perplexity: ~3x in about seven months

Why Anthropic Overtook

The most credible explanation is mix. Anthropic’s revenue is heavily weighted toward API and enterprise coding workloads — the token-hungry agentic use cases where a single customer can consume enormous volume without a corresponding seat count. Claude Opus 5, launched July 24, 2026 at $5/$25 per million tokens, sits directly in that lane, as does the Claude Code ecosystem.

OpenAI’s base is broader but historically more consumer-weighted, and consumer subscriptions have a ceiling per user that API consumption does not. Reporting in August 2026 indicated OpenAI’s enterprise revenue exceeded consumer for the first time, which is the more consequential fact in its own numbers.

An agentic coding customer’s spend scales with how much work the agent does. A chat subscriber’s spend is $20 a month regardless. That is the whole story of the gap.

Why the Run Rate Metric Deserves Suspicion

Annualized run rate takes a recent short period and multiplies it out. It is standard in private markets and it is systematically generous:

It flatters steep curves. A company growing 30% a month reports a run rate well above what it will actually collect over the next twelve months.
Enterprise contracts are lumpy. A few large deals landing in the measurement month inflate the annualization.
It ignores cost entirely. None of these figures say anything about profitability, and inference plus training costs at this scale are enormous.
None of it is audited. These are investor-update numbers relayed through journalists.

Anthropic’s figure carries additional weight because it was reportedly disclosed as part of a routine investor update ahead of an IPO process, where accuracy expectations are higher. That is a reason to take it more seriously than a typical leak — not a reason to treat it as audited.

Perplexity Is Measuring a Different Thing

Placing Perplexity’s $750 million next to Anthropic’s $65 billion invites the wrong conclusion. Perplexity buys inference — it committed $750 million over three years to Azure GPU capacity in January 2026 — and sells an application on top. Its cost of revenue is partly the model labs’ revenue.

Its growth is also arriving from an unexpected direction. Reporting attributes much of the 3x to Perplexity Computer, its cloud agent for professional desk work, rather than to search. That reframes the company: agent products monetize against labour budgets, which support far higher per-user pricing than an ad-adjacent search product.

Nvidia was reported on August 23, 2026 to be in talks to invest at a $30 billion-plus valuation, up from $21.21 billion earlier in the year — roughly 40x annualized revenue, which is aggressive even by 2026 standards and only defensible if the growth curve holds.

What This Means If You Are Building

Three practical takeaways:

The frontier price war is being funded by real revenue, not just capital. OpenAI cutting GPT-5.6 Sol to $4/$20 on August 21, 2026 and Anthropic cancelling its scheduled Sonnet 5 increase are moves by companies with money, which means cheap frontier inference is likely to persist rather than snap back.
Agentic workloads are where the money is, so that is where capability investment will keep going. Expect the fastest improvement in long-horizon coding and tool use.
Application-layer companies can reach real scale fast. Perplexity went from under $250 million to over $750 million in seven months without training a frontier model. The application layer is not a commodity.