GitHub Copilot has lost more ground in twelve months than most dominant developer tools lose in three years. The coding assistant that once commanded two-thirds of the market among professional developers now holds just 51% share, according to the Stack Overflow Developer Survey. Over the same period, Cursor, built by Anysphere, crossed $2 billion in annual recurring revenue, and Anthropic's Claude Code became the tool senior engineers trust most.

Why Is GitHub Copilot Losing Ground So Quickly?

The shift reflects a widening satisfaction gap between the incumbent and its challengers. When JetBrains surveyed senior engineers in April 2026 about which coding tool they loved most, Claude Code took 46% of the vote compared to just 9% for GitHub Copilot. That roughly 5-to-1 preference gap among the most experienced developers points to something raw usage numbers cannot capture: a fundamental trust problem.

Performance benchmarks tell a more nuanced story. GitHub Copilot scores 56% on SWE-Bench, a widely cited coding benchmark, against Cursor's 51.7%. On paper, Copilot's underlying model still performs well. However, Cursor completes comparable tasks roughly 30% faster by the same benchmark, and speed is what developers feel every single day inside their editor. A slightly higher accuracy score does not outweigh a tool that visibly slows people down.

The JetBrains Developer Ecosystem survey, which polled more than 10,000 developers in early 2026, found Copilot at 29% market share, with Cursor and Claude Code tied at 18% each. Neither survey shows Copilot losing its top spot outright, but the trend line is the real story. Almost all of Copilot's lost ground went to two rivals that barely registered in developer surveys three years ago.

How Are Developers Choosing Between Coding Assistants Today?

• Market Share Metrics: Stack Overflow data shows Copilot fell from 67% to 51% year over year, while JetBrains found Copilot at 29%, Cursor at 18%, and Claude Code at 18% among professional developers.
• User Satisfaction Rankings: Senior engineers prefer Claude Code over Copilot by roughly 5 to 1, with Claude Code capturing 46% of the "most-loved" vote compared to Copilot's 9%.
• Speed and Performance: Cursor completes tasks 30% faster than Copilot despite scoring slightly lower on accuracy benchmarks, making real-world responsiveness a decisive factor for developers.
• Pricing Models: All three major vendors have now adopted metered billing based on actual token usage, moving away from flat-rate subscriptions that proved unsustainable.

Cursor's climb to $2 billion in annual recurring revenue represents one of the fastest growth trajectories for any developer tool in software history. The company reported a paying user base above 1 million, though that specific figure has not been confirmed in an official disclosure. Cursor shipped version 0.45 on July 1, 2026, adding deeper support for background agents, which are AI tasks that keep running while a developer works on something else entirely.

Claude Code's rise has been the least expected part of this story. The tool reached 18% workplace adoption in a January 2026 reading, which JetBrains described as six-fold growth in just nine months. Among engineers who have actually used it, loyalty runs deep. Anthropic released Claude Sonnet 5 in July 2026 as its newest agentic-coding model, arriving days after Cursor's 0.45 release and underscoring how compressed release cycles have become across the entire category.

What Changed in GitHub's Billing Strategy?

GitHub's response to market pressure looks defensive in hindsight. On June 1, 2026, every GitHub Copilot plan switched from flat-rate premium requests to a metered system called AI Credits, priced at $0.01 per credit and consumed according to actual token usage. GitHub confirmed the change on its blog on April 27, 2026, five weeks ahead of the cutover, retiring the Premium Request Units system that had governed billing since 2025.

The timing is awkward for a company trying to win back share. Metered billing means heavy users, exactly the engineers most likely to compare Copilot against Cursor or Claude Code line by line, now see a bill that scales with how hard they push the tool. GitHub Copilot's shift to AI Credits arrived a full year after Cursor made the same change and about two weeks before Claude Code followed. Every major vendor has now reached the same conclusion: flat-rate AI coding tools were not sustainable to run, whatever it costs them in developer goodwill.

Are New Competitors Entering the Market?

Just as GitHub, Cursor, and Anthropic race to justify usage-based pricing, a new challenger is undercutting all three on cost. Z.ai, the Chinese AI lab formerly known as Zhipu AI, launched ZCode on July 2, 2026, a flat $16-per-month coding assistant running on the company's own GLM-5.2 model. That price sits well below what a heavy Cursor or Copilot user now pays once metered fees stack on top of a base subscription.

ZCode represents the first serious attempt by a Chinese lab to compete on price in this category rather than on benchmark leaderboards alone. It will not unseat the big three overnight, but it hands procurement teams a concrete number to point to in every renewal conversation. The launch landed in the same week as Cursor 0.45 and Claude Sonnet 5, underscoring how compressed release cycles have become across the whole category.

The broader market context makes these shifts significant. GitHub Copilot, Cursor, Claude Code, and a widening field of challengers now split a market estimated at $12.8 billion in 2026, up from $5.1 billion just two years earlier. For engineering leaders deciding which tool to standardize on, the figures tracking market share and user satisfaction matter more than marketing claims from any single vendor.