Anthropic's absent image model is the strategy

ANTHROPIC'S COMPETITORS have spent the past year selling the world on pictures. OpenAI's GPT Image 1, released in March 2025, produced roughly 700 million images for 130 million users in its first week, straining the company's GPUs to the point that Sam Altman publicly quipped they were "melting." Google's Nano Banana, launched five months later, drew 13 million new users in four days and had been used to generate more than five billion images by mid-October. xAI shipped Grok Imagine; ByteDance has Seedream; Meta has Imagine. Each of the frontier labs has lined up to cash in on the generative boom in consumer visuals. Each one, that is, except Anthropic.

Anthropic's own documentation states plainly that Claude "cannot generate, produce, edit, manipulate or create images," and leadership has repeatedly confirmed that image generation sits outside the near-term roadmap. The lab's multimodality is deliberately asymmetric: Claude can see — parse a screenshot, read a chart, interpret a UI — but it cannot produce a new pixel. In a market where every other frontier firm treats image generation as table stakes, the absence is conspicuous.

Drawing the line

But the gap is not an oversight. It is the cleanest statement of strategic focus any frontier lab has made. Anthropic's annualized revenue reached roughly $30bn by April, up from $9bn at the start of the year; the single largest engine of that growth was Claude Code, which by February had reached $2.5bn in run-rate revenue on its own. Roughly 70 to 75 percent of the company's revenue comes from API and enterprise contracts, and the monetization gap with OpenAI is stark — by one analyst estimate Anthropic earns around $211 per month per paying user against OpenAI's $25 per weekly active user, an eight-fold difference in revenue efficiency. Every H100 Anthropic owns is finite, and spending it on Ghibli portraits for TikTok is not what gets Pfizer, United Airlines, or Novo Nordisk to renew a seven-figure contract.