Anthropic vs. OpenAI: Anthropic Won Q2, but OpenAI Rebounded in Early Q3

On booked quarterly revenue, Anthropic therefore had the stronger result and overtook OpenAI for the first time in the reported comparison. The caveat is important: Anthropic's numbers were preliminary and came from documents seen by Bloomberg, while the companies do not provide identical public financial-reporting detail.

Profitability: a milestone for Anthropic, deeper losses for OpenAI

Anthropic reportedly recorded positive adjusted operating income in Q2, its first such milestone. “Adjusted” is the key qualifier: the figure should not automatically be interpreted as GAAP profitability, positive free cash flow or a fully profitable business.

OpenAI's Q2 result pointed in the opposite direction. The Wall Street Journal reported that revenue growth slowed while losses deepened, and SiliconANGLE, citing the same investor reporting, put OpenAI's operating loss at $12.3 billion, up from $9.3 billion in Q1.

The comparison gives Anthropic a stronger near-term profitability narrative, but it does not establish that its cost structure is permanently superior. Frontier-model economics can change quickly as training, inference, distribution and product-mix costs evolve.

Annualized revenue: Anthropic's reported run rate was higher

Anthropic's annualized revenue run rate reportedly surpassed $65 billion by late July, up from $47 billion in May, according to reporting based on figures shared with investors.

OpenAI offered a different but notable July signal. CFO Sarah Friar told employees that the company's July annualized recurring revenue exceeded its entire Q2 revenue. CNBC reported the statement from an internal meeting, while other coverage attributed the momentum to the GPT-5.6 model family, the ChatGPT Work enterprise agent and Codex.

These statements should not be read as saying that OpenAI recognized more revenue in July than it did during the whole quarter. Annualized recurring revenue is a run-rate calculation: it extrapolates a current pace over a year. Quarterly revenue is revenue recognized during a defined three-month period. Comparing the two directly can overstate the apparent difference.

The more defensible conclusion is that Anthropic had the larger reported run rate at the end of July, while OpenAI claimed a sharp improvement in its current sales pace after Q2.

Products: no universal winner is established

The product contest is harder to settle than the financial comparison. OpenAI launched the GPT-5.6 family—including Sol, Terra and Luna—across ChatGPT, Codex and its API in July. Secondary comparisons describe GPT-5.6 Sol as cheaper than Claude Fable 5 at the headline API price, while assigning different strengths to the models depending on the benchmark or workflow.

For example, available comparisons describe Fable 5 as competitive on difficult software-engineering tasks, while Sol is positioned more strongly for lower-cost agentic or terminal-oriented work. These are not equivalent tests, and several figures come from vendor-reported or secondary analyses rather than a single independent evaluation.

Early usage data provide a more practical signal than a single benchmark. The Decoder, citing Ramp data, reported that Fable 5 generated about 75% as much model-related spend as GPT-5.6 Sol in July, despite Fable 5's higher per-token price. That suggests Sol may have gained faster initial business adoption, but it does not prove that it was the better model overall or that OpenAI had won the broader product battle.

Enterprise adoption and API share: directionally clear, numerically unsettled

The available reporting supports a broad strategic distinction: Anthropic's growth appears heavily associated with enterprise and API demand, while OpenAI benefits from a broader consumer, developer and product ecosystem. Claude Code is frequently cited as an important contributor to Anthropic's momentum, but the specific claim that it had a $2.5 billion-plus annualized run rate is supported here mainly by lower-confidence industry estimates rather than a high-quality primary disclosure.

The same caution applies to precise market-share figures. A report cited a 32% enterprise API share for Anthropic versus 25% for OpenAI, but the methodology and underlying dataset are not sufficiently clear to treat those numbers as definitive market-share measurements.

Ramp-related adoption figures are also reported inconsistently across the available sources. Business Insider cited July shares of 43.5% of U.S. businesses paying for Anthropic products versus 39.7% for OpenAI, while the evidence supplied here does not include the underlying Ramp dataset or enough methodological detail to independently validate the comparison.

These numbers can be useful directional indicators, but they should not be presented as a complete measure of API market share, enterprise revenue or customer quality.

IPO prospects: stronger narrative for Anthropic, no confirmed timetable

Anthropic's reported revenue acceleration, positive adjusted operating income and $65 billion-plus run rate would strengthen the company's case for a future public offering. They would give potential investors a story built around rapid enterprise monetization and an earlier-than-expected path toward adjusted profitability.

However, the evidence provided does not reliably confirm the more specific claims that Anthropic confidentially filed a draft S-1 or retained Morgan Stanley, Goldman Sachs and JPMorgan as underwriters. Reports of IPO preparation should therefore be treated as indications of possible planning, not proof that a listing is committed, imminent or guaranteed at a particular valuation.

OpenAI also faces public-market pressure, with reporting describing a much-anticipated IPO and investor concern about its widening losses. Its July ARR statement could improve that narrative, but a stronger run rate does not by itself resolve questions about margins, capital intensity or the durability of post-launch demand.

What the comparison means for AI buyers and investors

Anthropic's Q2 numbers show that an enterprise-first strategy can scale rapidly when customers adopt models through APIs and workflow-specific tools. OpenAI's July update shows why a weak quarter does not necessarily translate into a durable loss of momentum: new models, enterprise products and coding tools can change the trajectory quickly.

The most useful takeaway is not that one company has permanently won. Enterprise spending can move toward the model that offers the best combination of quality, reliability, price and workflow fit. But low switching costs do not make platform advantages irrelevant. Distribution, integrations, developer tools, security controls, compliance capabilities and embedded workflows can still influence retention.

Bottom line: Anthropic won the reported Q2 comparison on revenue growth and adjusted operating performance. OpenAI's strongest evidence came from early-Q3 momentum rather than its quarter-end financial profile. The leadership race remained unsettled, and the most precise claims about API share, product adoption and IPO banking arrangements required more reliable corroboration than the available evidence supplied.