Anthropic Passes OpenAI in Enterprise; Coding Agent War Ignites
For the first time, Anthropic has overtaken OpenAI in enterprise AI adoption. Menlo Ventures' April 2026 survey records Anthropic at 34.4% (+3.8 points) against OpenAI's 32.3% (−2.9 points) — a narrow but symbolically significant inversion. Within an hour of the data surfacing, both companies launched competing counter-offers that reveal the competitive stakes more clearly than any benchmark.
What the Source Actually Says
Nate Herk's May 13 analysis documents the response timeline precisely: Sam Altman tweeted that Codex is "the best AI coding product" and offered enterprise customers two free months of Codex usage if they switch from Claude Code within 30 days. Approximately 45 minutes later, Anthropic's @ClaudeDevs account announced a 50% increase in Claude Code weekly limits through July 13 — exactly two months out. Three independent source batches corroborated both the enterprise flip and the competitive responses within the same reporting window.
Herk's read: both labs are deliberately operating in a "free-sample phase," pricing coding agents far below their real value. He estimates that $200/month of Claude Code or Codex usage delivers the equivalent of $5,000–$15,000 in engineering output — a subsidy ratio of 25–75×. The motivation is not charity: labs need proprietary usage patterns and habit-forming adoption more urgently than subscription revenue. Altman's own prior admission that OpenAI was losing money on Pro subscriptions because users "dramatically over-consumed vs expectation" confirms the unit economics are intentional.
Independent Vercel Gateway usage data published the same day adds texture: Anthropic already leads Vercel traffic for vibe-coding and back-office categories; OpenAI leads in recruiting outreach. The enterprise flip is not uniform — it is a category-by-category contest still very much in motion.
Strategic Take
July 13 is the key date: Claude Code's 50% limit boost expires precisely when Codex's two-month trial window closes. The synchronized end-date is almost certainly not coincidental. Build agent workflows that are harness-portable now — the free-sample window is finite, and both labs have structural incentives to normalize pricing once adoption and data collection targets are met.
