A year ago, OpenAI owned the AI conversation. Now, in the category that might matter most to its business - AI-powered coding - it's chasing Anthropic.
That's the uncomfortable reality behind the current state of the AI coding tools market. Claude Code launched in February 2025, and for months it was a capable but under-hyped product. Then Anthropic shipped the Opus 4.5 model in November 2025, and everything changed. One industry researcher put it bluntly: Claude Code had been around for a year, but it didn't get this level of attention because the models simply weren't good enough. Opus 4.5 fixed that.
Since then, Anthropic's numbers have been staggering. Annualized revenue doubled to $2.5 billion. Active users doubled. Business subscriptions quadrupled. Enterprise customers now represent over half of revenue. The company's valuation sits at $380 billion.
The Benchmark Gap That Isn't
On paper, OpenAI's Codex and Claude Code are neck and neck. On SWE-bench Verified - a standardized test that measures how well AI can solve real software engineering problems - Claude Code (Opus 4.6) scores 80.9% versus Codex at roughly 80%. Codex actually wins on Terminal-Bench 2.0, which tests autonomous terminal operations, at 77.3%.
But benchmarks don't explain adoption. Developers who use both tools describe a clear difference in feel: Claude Code responds quickly and concisely, functioning as an autonomous coding agent that handles entire projects from the terminal. Codex tends to deliberate longer, producing more detailed but slower answers. For the rapid iteration cycles that define modern development, speed wins.
OpenAI has been shipping aggressively to close the gap. GPT-5.3-Codex arrived in February with a claimed 25% speed improvement. Both companies launched competing security scanning products in early March - Codex Security and Claude Code Security - within weeks of each other.
Developer Pressure Is Real
The stakes here go beyond bragging rights. Organizations are adopting these tools fast, driven by developer demand from the ground up. As one cybersecurity researcher noted, companies that fail to adopt AI coding tools risk falling behind. That pressure creates a winner-take-most dynamic. Developers pick a tool, build muscle memory around it, and switching costs climb.
Google's Gemini CLI offers a free tier and competitive models, but the consensus among analysts is that Anthropic and OpenAI are currently stronger for coding tasks. The real race is between two.
Price Is the Wildcard
Both tools start at around $20 per month, with advanced tiers reaching $200. But heavy users hit token-based pricing quickly, and Claude Code tokens are considered particularly expensive. If OpenAI can match Claude Code's quality while undercutting on price - something it has the resources to do - the market could shift. For now, though, Anthropic has something OpenAI rarely concedes to anyone: a head start and momentum.