AI Coding Assistant Market Share: Claude Code Overtakes Copilot

The leader has changed

For most of the short history of AI coding tools, GitHub Copilot was the default name. A new market model published in October 2026 says that is no longer true. First Page Sage surveyed 2,350 professional developers in North America and Europe between August and September, then adjusted the answers against disclosed seat counts and usage figures. Its model gives Anthropic's Claude Code 29.4% of the market as a developer's main coding assistant. That is up 10.9 points since January. GitHub Copilot fell 7.8 points to 22.7%, Cursor dropped to 13.1%, and OpenAI Codex rose to 11.8%41. In the firm's quarterly figures, Claude Code passed Copilot between Q2 and Q3 202641.

Large survey data points the same way. In JetBrains' 2026 fieldwork with more than 15,000 developers, 39% said they use Claude Code at work, up from 18% in January. The figure is 47% in the United States. Copilot came second as a main tool at 21%, ahead of Codex at 16% and Cursor at 12%50. The two methods are very different, but they agree on the order at the top. Our reading is that Claude Code now leads among individual developers. That is a different claim from leading the whole market.

Why the numbers disagree

The gap between headline figures is mostly about what each one measures. A figure saying Claude Code holds 54% of the AI coding market is widely repeated, and it is misleading. It comes from Menlo Ventures and measures Anthropic's share of enterprise coding models, not how many developers use Claude Code4640. Some coverage still presents it as Claude Code's own market share31.

The model data explains why the two get confused. First Page Sage estimates that Claude models produced 47.3% of AI-written code in its sample. Claude powers Claude Code, and developers also choose it inside Cursor, Copilot and most open-source agents41. Anthropic earns from its rivals' success as well as its own.

Revenue figures need the same care. The last Claude Code revenue figure Anthropic itself released was a run rate above $2.5 billion in February 2026. At that point, weekly active users had doubled since January 1 and business subscriptions had quadrupled3336. Higher numbers, about $8 billion in May and close to $15 billion by August, come from third-party trackers rather than the company33. Some articles report the $8 billion as settled fact31. It should be treated as an estimate. Either way, Claude Code sits inside a much larger business: Anthropic reported a total run rate above $47 billion by late May33.

Copilot: still the biggest, but under pressure

Copilot's slide in share sits alongside growth in absolute numbers. Satya Nadella said in July that Copilot had 50 million users. That count includes free, student and paid accounts, so it shows reach more than revenue11. Microsoft reported 4.7 million paid subscribers in January 2026, up 75% from a year earlier1216. Microsoft also says one in three pull requests on GitHub now involves an AI agent1149.

The bigger story this year is pricing. On June 1, every Copilot plan switched from premium request units to usage-based GitHub AI Credits, priced by tokens. Base prices stayed at $10, $39, $19 and $39 for the Pro, Pro+, Business and Enterprise tiers13. GitHub said the old model, in which a quick chat and a multi-hour autonomous session could cost the same, was no longer sustainable19. Ars Technica cited reported internal documents showing Copilot's week-over-week costs had nearly doubled since January19.

Users pushed back. In the weeks before the switch, GitHub paused new sign-ups for Pro, Pro+ and Student plans, tightened usage limits, and removed Opus models from Pro14. Once billing went live, Business Insider and PCMag reported users burning through half their monthly credits in a day or two. Some posted cost estimates in the hundreds or even thousands of dollars a month1720. Visual Studio Magazine described a common complaint: users get less for the same price, and costs are harder to predict15.

In our view, this reflects the economics of the whole category more than a mistake by one vendor. Agent workflows use far more compute than autocomplete. Flat-rate plans hid that cost for a while. PCMag noted that Anthropic moved Claude Enterprise to token billing in April, before GitHub did20.

Copilot's advantage is in large companies. First Page Sage puts Copilot at 35.6% among firms with more than 5,000 employees, where existing Microsoft contracts make it the easy purchase. Claude Code has 22.4% in that group, while it leads startups with 36.8%41. Other trackers also show Copilot ahead in large enterprises, though their figures differ4218. Developer satisfaction surveys are less kind: in the JetBrains survey cited by IdeaPlan, 9% named Copilot their most-loved tool, against 46% for Claude Code42.

Cursor: falling share, rising revenue

Cursor shows clearly that market share and revenue are different things. Its share of primary-tool usage peaked in Q4 2025 and has fallen every quarter since, according to First Page Sage41. Over the same period its revenue grew very fast. Dealroom reported annualized revenue above $4 billion in early June, up from $3 billion in late April and $2 billion in February. About 75% of that came from enterprise customers5.

That growth drew a buyer. SpaceX agreed on June 16 to buy Anysphere, Cursor's parent company, in an all-stock deal worth $60 billion. The deal was announced days after SpaceX's own IPO110 and closed on August 142. It followed an April option agreement under which SpaceX could buy Cursor for $60 billion or walk away after paying about $10 billion2. Cursor now sits in a SpaceXAI division alongside Grok2.

The reported numbers vary somewhat. One tracker puts Cursor at 26% market share, about double First Page Sage's figure1. Another reports that the company still had negative gross margins while it pursued a target of more than $6 billion in ARR6. Our reading is that Cursor won over paying teams but is losing ground to tools that work from the terminal. The inference-cost problem facing Copilot affects Cursor too, because it pays for models it does not own.

The shift toward agents

The main change behind these rankings is in how developers work, not which brand they pick. First Page Sage estimates that terminal and command-line agents grew from 21.3% of primary usage in January to 38.6% in October. IDE extensions fell from 41.2% to 29.4%, and AI-native editors also declined41. That shift helps the agent-first tools, Claude Code and Codex. It also explains why Microsoft says usage of Copilot CLI, its own terminal tool, was nearly doubling month over month in April11. OpenAI's Codex, the year's other big gainer, roughly tripled its share as developers adopted its cloud task mode41.

The market is also concentrating. The four biggest tools held 77% of usage in October, up from 71.2% in January41. Sizing the market as a whole is still unreliable: 2026 estimates run from $4.1 billion to $16.13 billion, depending on what each firm counts43.

Code review becomes the next contest

When agents write more of the code, the bottleneck becomes checking it. AI code review is now a busy market of its own. Copilot's built-in reviewer wins on convenience: comparisons recommend it for teams already paying for Copilot on GitHub2329. Since June, though, Copilot reviews use AI Credits and GitHub Actions minutes, so a feature that felt included is now metered1329.

Specialist tools compete on review quality. CodeRabbit is usually described as the low-noise option that works on all major Git platforms, starting at $24 per user per month2423. Greptile indexes the entire codebase and finds more cross-file bugs, but raises more false alarms24. Qodo targets companies that need to enforce rules and compliance30. Anthropic has its own Claude Code Review. TechCrunch reported average costs of $15 to $25 per review, and at launch it was a research preview for Teams and Enterprise customers29.

Reviewers disagree on how much context Copilot's reviewer sees. Some describe it as diff-only23. Others say it can draw on the full project, repository instructions and MCP servers30. Our reading is that the product has moved faster than many reviews of it.

Our reading

Claude Code is the momentum leader, and Anthropic gains even when developers pick a rival tool, because rivals often run Claude models. Copilot still has distribution and enterprise contracts, but its pricing change reveals the cost of agent-style use, and it lost goodwill making that change. Cursor proved there is a premium market for an AI-native editor and sold at the high point. Over the next year, the deciding factor looks likely to be who can run agents cheaply enough to make money on them.