AI Coding Assistant Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: November 5, 2026 | License: CC BY 4.0
Claude Code is now used at work by 39% of professional developers worldwide against 21% for GitHub Copilot, per JetBrains’ May–July 2026 survey. Axis Intelligence Research’s Incumbent Crossover Ratio (ICR™) puts the challenger’s lead at 1.86x, up from 0.10x in April–June 2025. Stack Overflow’s 2026 survey shows the same flip.
Quick Answer
65.9% of developers use an AI coding assistant or agent at work (Stack Overflow 2026); 90% of professional developers use one weekly (JetBrains). Agents started over 40% of GitHub pull requests in September 2026. Copilot has 50 million users, but Claude Code leads work use. Axis Intelligence Research annualizes Cursor’s filed revenue at $4.62 billion.
Key Findings
- Axis Intelligence Research finds the AI coding market changed leaders in 2026: the Incumbent Crossover Ratio (Claude Code usage ÷ GitHub Copilot usage) rose from 0.10 to 1.86 across four JetBrains waves (April 2025 to July 2026) and from 0.60 to 1.12 in Stack Overflow data; on stated intent for the next six months it reaches 2.11.
- Axis Intelligence Research estimates Cursor’s revenue run-rate at about $4.62 billion, annualizing the $608 million SpaceX reported for Cursor between August 14 and September 30, 2026.
- Agents started more than 40% of GitHub pull requests in September 2026, up from 17% in March, per GitHub’s Octoverse 2026; at that pace, Axis Intelligence Research estimates agents crossed half of new pull requests around mid-December 2026.
- Axis Intelligence Research calculates that 65.7% of developers switched AI tools in the past year, and that quality-driven switching outnumbers price-driven switching 2.59 to 1, using Stack Overflow’s 2026 data (n=13,144).
- AI coding prices now run on a 10x to 25x ladder: Axis Intelligence Research’s Power Tier Multiple shows the top individual plan costs 10x the entry plan at GitHub Copilot and Claude, at least 20x at ChatGPT and 25x at Cursor, as of November 5, 2026.
How Many Developers Use AI Coding Assistants in 2026?
Two of every three developers use an AI coding assistant or coding agent in their job, and the habit is daily. In Stack Overflow’s 2026 Developer Survey, 65.9% of 17,464 respondents said they currently use “AI coding assistants or coding agents” at work, while 17.2% said they use no AI tools at all. Among developers who use coding tools, 73.0% use them every day, and 30.9% of AI users spend four or more hours a day with them.
JetBrains measures professional developers only and finds deeper penetration: 90% use AI coding agents at work at least weekly and 68% use them daily, according to its Developer Ecosystem Survey 2026 agent-adoption report (published August 18, 2026; 15,509 respondents, May–July 2026).
| Measure | Value | Population | Source (date) |
|---|---|---|---|
| Use AI coding assistants or agents at work | 65.9% | All respondents, n=17,464 | Stack Overflow 2026 (Oct 6, 2026) |
| Use no AI tools | 17.2% | All respondents, n=17,464 | Stack Overflow 2026 |
| Use coding assistants/agents daily | 73.0% | Users of those tools, n=14,478 | Stack Overflow 2026 |
| 4+ hours a day with AI tools | 30.9% | AI users, n=10,926 | Stack Overflow 2026 |
| Use AI coding agents at least weekly | 90% | Professional developers, n=15,509 | JetBrains (May–Jul 2026) |
| Use AI coding agents daily | 68% | Professional developers | JetBrains (May–Jul 2026) |
| Use 2+ AI tools in a typical week | 79.7% | AI users, n=11,919 | Axis calculation on Stack Overflow 2026 |
The two surveys disagree on the headline only because they ask different people different questions. Stack Overflow’s base includes students, hobbyists and developers at organizations that do not use AI for work (6.1% of respondents); JetBrains reweights a professional sample by region, employment status and programming language. A 65.9% floor and a 90% ceiling is the honest range for 2026.
Sentiment followed usage up. In the same Stack Overflow data, 62.0% of respondents hold a favorable view of AI tools and 21.5% an unfavorable one, a ratio of 2.88 to 1. Trust is conditional rather than broad: 48.0% trust AI output only when they can verify it easily, and just 6.6% trust it for important decisions.
Sarah Mitchell: The adoption question is closed. What the 2026 data actually measures now is hours, not headcount — almost a third of AI users spend half a workday inside these tools. That is the number vendors will price against in 2027, and it is why per-seat pricing is disappearing.
For workflow-stage detail (planning, review, deployment), see our developer AI adoption statistics.
Which AI Coding Assistant Do Developers Use Most in 2026?
Claude Code is the most-used AI coding tool in both major 2026 developer surveys, ahead of GitHub Copilot. Stack Overflow asked 12,255 developers which coding agents or assistants they used in the past year: Claude Code 65.5%, GitHub Copilot 58.7%, OpenAI Codex 29.5% and Cursor 25.5% (Stack Overflow 2026, AI coding agents).
AI coding tool usage, Stack Overflow 2026 vs 2025
| Tool | 2026: used in past year (n=12,255) | 2025: share of AI agent users (n=8,323) | Source |
|---|---|---|---|
| Claude Code | 65.5% | 40.8% | Stack Overflow 2026; Stack Overflow 2025 |
| GitHub Copilot | 58.7% | 67.9% | Stack Overflow 2026; Stack Overflow 2025 |
| OpenAI Codex | 29.5% | — | Stack Overflow 2026 |
| Cursor | 25.5% | — | Stack Overflow 2026 |
| Google Antigravity | 16.0% | — | Stack Overflow 2026 |
| Gemini Code Assist | 14.6% | — | Stack Overflow 2026 |
| OpenCode | 13.6% | — | Stack Overflow 2026 |
| JetBrains AI | 12.8% | — | Stack Overflow 2026 |
| Windsurf | 6.5% | — | Stack Overflow 2026 |
| Amazon Q Developer | 3.9% | — | Stack Overflow 2026 |
| Devin | 2.9% | — | Stack Overflow 2026 |
| Tabnine | 2.1% | — | Stack Overflow 2026 |
The 2025 base (respondents who use or build AI agents) is narrower than the 2026 base, so the year-over-year comparison shows direction, not exact change.
AI coding tool work adoption, four JetBrains waves (April 2025 – July 2026)
| Tool | Apr–Jun 2025 (n=16,714) | Sep 2025 (n=5,763) | Jan 2026 (n=10,739) | May–Jul 2026 (n=15,509) | Source |
|---|---|---|---|---|---|
| Claude Code | 3% | 12% | 18% | 39% (US: 47%) | JetBrains |
| GitHub Copilot | 29% | 31% | 29% | 21% | JetBrains |
| OpenAI Codex | 0% | 1% | 3% | 16% | JetBrains |
| Cursor | 12% | 16% | 18% | 12% | JetBrains |
| JetBrains AI Assistant and/or Junie | 13% | 11% | 11% | 9% | JetBrains |
| OpenCode | — | — | — | 7% | JetBrains |
| Google Antigravity | — | — | 6% | 6% | JetBrains |
Claude Code is the single most-used AI coding tool for 31% of professional developers, which means 79.5% of its regular work users treat it as their main tool (Axis calculation: 31 ÷ 39). Codex grew 5.33x between January and May–July, from 3% to 16%. Cursor fell from 18% to 12%, with the steepest drop in China (28% to 16%).
The Incumbent Crossover Ratio (ICR™)
The Incumbent Crossover Ratio (ICR™) is an Axis Intelligence Research metric that tracks when an incumbent coding assistant loses its lead. It divides the usage share of the leading agent-first challenger by GitHub Copilot’s usage share in the same survey wave, so each reading stays inside one methodology. Above 1.0, the incumbent has been overtaken.
ICR = challenger usage share ÷ GitHub Copilot usage share (same survey, same wave)
| Survey wave | Claude Code | GitHub Copilot | ICR™ |
|---|---|---|---|
| JetBrains DevEco, April–June 2025 (work use) | 3% | 29% | 0.10 |
| Stack Overflow 2025 (AI agent users) | 40.8% | 67.9% | 0.60 |
| JetBrains AI Pulse, September 2025 (work use) | 12% | 31% | 0.39 |
| JetBrains AI Pulse, January 2026 (work use) | 18% | 29% | 0.62 |
| Stack Overflow 2026 (past-year use) | 65.5% | 58.7% | 1.12 |
| JetBrains DevEco, May–July 2026 (work use) | 39% | 21% | 1.86 |
| Stack Overflow 2026, want to use in next 6 months (Forward ICR™) | 41.4% | 19.6% | 2.11 |
Two independent surveys, with different samples and different questions, both cross 1.0 in 2026. In the JetBrains series the crossover happened between January and May–July 2026, after the ratio climbed from 0.10 through 0.39 and 0.62. The gap between 1.12 (Stack Overflow, any use in the past year) and 1.86 (JetBrains, regular work use) is informative too: Copilot keeps far more casual users than committed ones, which is what a bundled, widely licensed incumbent looks like when developers choose their own daily tool. The Forward ICR™ applies the same formula to stated intent, and at 2.11 it says the gap is still widening.
Which AI coding tools do developers want to use next?
Stack Overflow asked the same 12,255 developers which coding agents they want to use in the next six months (Stack Overflow 2026, “Want to use” view of the same question). Axis Intelligence Research divides each tool’s “want to use” count by its “used in the past year” count to get an intent-to-usage ratio: below 1.0, fewer developers want the tool than already use it.
| Tool | Want to use, next 6 months | Used, past year | Intent-to-usage ratio (Axis) | Source |
|---|---|---|---|---|
| Claude Code | 41.4% | 65.5% | 0.63 | Stack Overflow 2026 |
| GitHub Copilot | 19.6% | 58.7% | 0.33 | Stack Overflow 2026 |
| OpenAI Codex | 17.6% | 29.5% | 0.59 | Stack Overflow 2026 |
| Cursor | 11.7% | 25.5% | 0.46 | Stack Overflow 2026 |
| OpenCode | 11.7% | 13.6% | 0.86 | Stack Overflow 2026 |
| Google Antigravity | 9.1% | 16.0% | 0.57 | Stack Overflow 2026 |
| Gemini Code Assist | 7.0% | 14.6% | 0.48 | Stack Overflow 2026 |
| JetBrains AI | 6.5% | 12.8% | 0.50 | Stack Overflow 2026 |
| Pi | 6.6% | 4.4% | 1.52 | Stack Overflow 2026 |
| Mistral Vibe | 4.6% | 3.2% | 1.43 | Stack Overflow 2026 |
| Windsurf | 3.4% | 6.5% | 0.52 | Stack Overflow 2026 |
GitHub Copilot has the weakest pull of any major tool: for every three developers who used it in the past year, one wants to use it next. Only two tools draw more intent than usage, Pi (1.52) and Mistral Vibe (1.43). Both are small, which makes them the ones to watch in the first 2027 survey waves. Among established tools, the open-source OpenCode holds intent best (0.86).
Awareness-to-work conversion
Dividing JetBrains’ work-use share by its awareness share shows how many developers who know a tool actually use it at work. In May–July 2026, Claude Code and GitHub Copilot were equally well known (79% awareness each), but Claude Code converted 49.4% of the developers aware of it into work users against 26.6% for Copilot. In January 2026 the order was reversed: 31.6% for Claude Code, 38.2% for Copilot. Codex converted 24.6%, OpenCode 16.7%, Cursor 16.0% and Google Antigravity 12.8%.
| Tool | Awareness Apr–Jun 2025 | Awareness Jan 2026 | Awareness May–Jul 2026 | Conversion May–Jul 2026 (Axis) | Source |
|---|---|---|---|---|---|
| Claude Code | 31% | 57% | 79% | 49.4% | JetBrains |
| GitHub Copilot | 79% | 76% | 79% | 26.6% | JetBrains |
| Cursor | 61% | 69% | 75% | 16.0% | JetBrains |
| OpenAI Codex | 11% | 27% | 65% | 24.6% | JetBrains |
| JetBrains AI Assistant and/or Junie | 57% | 50% | 53% | — | JetBrains |
Sarah Mitchell: Copilot’s problem isn’t reach. Exactly as many professional developers know it as know Claude Code. Its problem is that knowing it no longer turns into using it. The 2025 playbook of bundling an assistant into the platform worked until developers could compare agents side by side. The ICR says they did, and they chose.
Per-tool deep dives: GitHub Copilot statistics, Claude AI statistics, Cursor AI statistics and the head-to-head GitHub Copilot vs Cursor statistics.
How Much Code Do AI Agents Write on GitHub?
AI agents started more than 40% of pull requests on GitHub in September 2026, up from 17% in March, according to GitHub’s Octoverse 2026 report, which covers October 1, 2025 to September 30, 2026. Agent-started pull requests reached 110 million in September alone, about six times March’s volume.
| GitHub metric | Value | Period | Source |
|---|---|---|---|
| Developers on GitHub | 225M+ | Sept 2026 | GitHub Octoverse 2026 |
| Agent-started pull requests (share) | 40%+ | Sept 2026 | GitHub Octoverse 2026 |
| Agent-started pull requests (share) | 17% | March 2026 | GitHub Octoverse 2026 |
| Agent-started pull requests (volume) | 110M | Sept 2026 | GitHub Octoverse 2026 |
| Agent PRs merged within two weeks | 90% (Feb: 83%) | Sept 2026 | GitHub Octoverse 2026 |
| Non-agent PRs merged within two weeks | about 86% | Sept 2026 | GitHub Octoverse 2026 |
| Commits | 34.3B (2025: 12.9B) | 12 months to Sept 2026 | GitHub Octoverse 2026 |
| Merged pull requests | 1.3B (2025: 538M) | 12 months to Sept 2026 | GitHub Octoverse 2026 |
| Copilot MAU using agentic features | 55% (year earlier: 35%) | Sept 2026 | GitHub Octoverse 2026 |
Commits grew 2.66x in a year (Axis calculation: 34.3 ÷ 12.9). Merge quality did not fall with volume: agent-started pull requests now merge within two weeks at a higher rate (90%) than pull requests without an agent signal (about 86%).
AI involvement in merged pull requests by language
| Language | Merged PRs with AI involvement | Source |
|---|---|---|
| TypeScript | 44.1% | GitHub Octoverse 2026 |
| Python | 41.1% | GitHub Octoverse 2026 |
| Rust | 40.4% | GitHub Octoverse 2026 |
| Java | 21.2% | GitHub Octoverse 2026 |
Why “40%” and “one in three” are both right
Microsoft’s CEO said on July 29, 2026 that “one in three pull requests on GitHub now involves an agent” (Microsoft FY26 Q4 earnings call), and on October 28 that agents “open more than a third of all PRs.” GitHub’s own report says more than 40% in September. The figures measure different things at different dates: Octoverse counts pull requests started with an agent signal in a single month; the earnings-call phrasing is a rounded platform-wide statement. Axis Intelligence Research uses the Octoverse series because it publishes its method and its monthly path.
Claude Code alone authored 4% of all public GitHub commits as of February 12, 2026, per Anthropic. Inside Anthropic, the company’s S-1 filing states that over 90% of contributions to its codebase in September 2026 were written by Claude. It adds that the average daily volume of code contributed per engineer in the quarter ended September 30, 2026 was over 45 times the 2024 average. Language-level detail lives in our programming language statistics and platform data in our GitHub statistics.
Sarah Mitchell: The merge-rate number is the one to watch. A year ago the objection to agents was that their pull requests rotted in review. GitHub’s data now shows agent PRs merging faster than the human baseline. Either agents got better, or teams stopped reading what they merge — and Octoverse can’t tell you which.
How Big Is the AI Coding Assistant Market? Users and Revenue by Vendor
No vendor publishes a clean “AI coding revenue” line, so the market has to be assembled from filings and company statements. The table below labels every figure by its source type.
| Vendor / product | Latest figure | Type | As of | Source |
|---|---|---|---|---|
| GitHub Copilot | 50M users | Company-reported | Jul 29, 2026 | Microsoft FY26 Q4 call |
| GitHub Copilot | 4.7M+ paid subscribers (+75% YoY) | Company-reported | Jan 28, 2026 | Microsoft FY26 Q2 call |
| GitHub Copilot | Nearly 140,000 organizations | Company-reported | Apr 29, 2026 | Microsoft FY26 Q3 call |
| Claude Code | $2.5B+ run-rate revenue | Company-reported | Feb 12, 2026 | Anthropic |
| Cursor | $608M revenue, Aug 14–Sep 30, 2026 | Filed (consolidated by SpaceX) | Sep 30, 2026 | SpaceX Q3 2026 results |
| Cursor | ~$4.62B annualized | Axis Intelligence Research estimate | Sep 30, 2026 | Axis calculation |
| OpenAI Codex | 5M+ weekly active users | Company-reported | Jun 2, 2026 | OpenAI |
| JetBrains Junie | 343,000 users | Company-reported | 2025 | JetBrains Annual Highlights 2026 |
GitHub Copilot
Microsoft reported 50 million GitHub Copilot users on its July 29, 2026 call, and “over 4.7 million paid Copilot subscribers, up 75% year-over-year” on January 28, 2026. Do not divide one by the other: the two counts measure different populations six months apart. Microsoft has not disclosed a paid count since, and on October 28, 2026 cited “stronger GitHub Copilot adoption and usage” rather than a new number.
Claude Code
Anthropic said on February 12, 2026 that Claude Code’s run-rate revenue had “grown to over $2.5 billion,” against total company run-rate revenue of $14 billion (Anthropic Series G announcement). That made Claude Code at least 17.9% of Anthropic’s run-rate (Axis calculation). Anthropic’s Form S-1, filed October 26, 2026, does not break out Claude Code revenue; it states that the top 5% of Claude Enterprise users who used Claude Code spent approximately 20 times the monthly amount of Enterprise users who did not use it, in the quarter ended September 30, 2026.
Cursor
Cursor’s first revenue figure in a public-company filing came from its acquirer. SpaceX completed the acquisition on August 14, 2026 at an implied equity value of $60 billion (GAAP purchase consideration: $42.5 billion), and reported that “from the closing date through September 30, 2026, Cursor contributed $608 million of revenue” (SpaceX Q3 2026 results).
Axis Intelligence Research calculation: $608 million × 365 ÷ 48 days (August 14 to September 30, inclusive) = $4.62 billion annualized. This is a straight-line estimate on 48 days of recognized revenue, not a disclosed run-rate. For context, Bloomberg reported on March 2, 2026, citing a person familiar with the matter, that Cursor’s annualized revenue topped $2 billion in February. The filed figure implies a pace 2.31 times higher by the third quarter.
OpenAI Codex
OpenAI reported “more than 5 million weekly active users” for Codex on June 2, 2026, up more than 6x since its desktop app launched in February, with knowledge workers about 20% of users (OpenAI). On October 7, 2026, Codex lead Tibo Sottiaux posted “a new high of 40M active users across Codex and ChatGPT Work”. That figure combines two products, so it can’t be compared with the Codex-only count.
Sarah Mitchell: Strip away the valuations and two things are filed: Anthropic’s S-1 and SpaceX’s consolidated revenue. Everything else in this market is company-reported or estimated. Analysts quoting a single “AI coding market size” for 2026 are adding numbers that were never designed to be added.
How Much Do AI Coding Assistants Cost in 2026?
Every major AI coding assistant now prices on usage, and individual paid plans span $5 to $200 a month. GitHub moved all Copilot plans to usage-based billing on June 1, 2026, replacing premium requests with GitHub AI Credits consumed at model API rates; code completions do not consume credits.
AI coding assistant prices (verified November 5, 2026)
| Vendor | Entry paid plan | Mid plan | Top individual plan | Team/business seat | Source |
|---|---|---|---|---|---|
| GitHub Copilot | Pro $10/mo | Pro+ $39/mo | Max $100/mo | Business $19, Enterprise $39 per user/mo (credits included) | GitHub Copilot plans |
| Cursor | Plus $8/mo | Premium $30/mo, Super $100/mo | Ultra $200/mo | Teams from $30 per user/mo | Cursor pricing |
| Claude (Claude Code included) | Pro $20/mo (monthly billing) | Max $100/mo | Ultra $200/mo | Team $25 standard / $125 premium per seat/mo (monthly billing) | Claude pricing |
| ChatGPT (Codex included) | Go $5/mo | Plus $20/mo | Pro from $100/mo (3 usage levels) | Business: per-user, price not listed | ChatGPT pricing |
Prices exclude tax. Cursor’s $8 Plus plan uses a shared pool limited to Grok models, and third-party models in Cursor are billed separately at API rates. ChatGPT Free and Go include limited Codex access, Plus includes extended Codex usage, and Pro adds longer Codex sessions plus up to $500 a month in API credits.
Power Tier Multiple
The Power Tier Multiple (PTM) is an Axis Intelligence Research calculation: the highest self-serve individual price divided by the lowest paid individual price.
| Vendor | Top tier ÷ entry tier | PTM |
|---|---|---|
| GitHub Copilot | $100 ÷ $10 | 10x |
| Claude | $200 ÷ $20 | 10x |
| Cursor | $200 ÷ $8 | 25x |
| ChatGPT (Codex) | $100+ ÷ $5 | 20x or more |
A 10x–25x ladder is what consumption pricing looks like when it is packaged as subscriptions. Microsoft said on October 28, 2026 that credit consumption from agentic features on GitHub “has more than doubled since usage-based billing went into effect.” Budgeting guidance for the underlying model costs is in our Claude API pricing and OpenAI API pricing pages.
Sarah Mitchell: The seat is now a minimum charge, not a price. Budget owners comparing $19 against $20 are reading the wrong line — the bill is set by how many agent runs a team launches.
Why Do Developers Switch AI Coding Tools?
About two in three developers switched AI tools in the past year, and quality drove it far more than price. Axis Intelligence Research calculates a 65.7% switching rate from Stack Overflow’s 2026 data (100% minus 33.7% who did not switch, minus 0.6% not applicable; n=13,144).
| Main reason for switching | Share of respondents | Source |
|---|---|---|
| Better output quality | 25.1% | Stack Overflow 2026 |
| Organization required or recommended a tool | 10.6% | Stack Overflow 2026 |
| Price-driven (lower cost 6.4% + previous tool too expensive or limited 3.3%) | 9.7% | Axis calculation on Stack Overflow 2026 |
| Did not switch | 33.7% | Stack Overflow 2026 |
Quality-driven switching outnumbers price-driven switching 2.59 to 1 (25.1 ÷ 9.7). Employers are tightening the choice: 21.7% of respondents say AI is expected for certain work (11.5%) or measured as part of performance (10.2%).
Sarah Mitchell: This is the least loyal software category developers have ever bought. When two-thirds move in a year and price is a minor reason, the moat is the model behind the tool — which is exactly why GitHub now routes across dozens of models and no single model carries more than 25% of Copilot requests.
Do AI Coding Assistants Make Developers Faster?
The best controlled evidence is still inconclusive on size, though the direction has turned positive. METR’s early-2025 randomized trial (16 experienced open-source developers, 246 tasks) found AI made tasks take 19% longer. Its February 2026 update (57 developers) estimated an 18% speedup for returning developers and 4% for new ones, with confidence intervals that include zero. METR also reports that 30% to 50% of surveyed developers skipped some tasks because they did not want to work without AI, which biases the estimates downward.
Workflow-stage productivity data is in our developer AI adoption statistics, and delivery-pipeline effects in our DevOps statistics.
Is Code From AI Coding Assistants Secure?
AI-generated code passed Veracode’s security tests 62% of the time in October 2026, the first meaningful gain since 2025. Veracode tested 18 models across Python, C#, JavaScript and Java; the average rose from 56% in its July 2026 report, and the best models (Claude Opus 5.5 and GPT-6-Sol) scored 73%. Put the other way, 38% of generated samples still failed.
Stack Overflow’s data matches that caution: 19.9% of developers use AI for deploying or operating production systems, against 69.3% for generating code in a familiar area. Benchmark breakdowns by vulnerability class are in our AI-generated code statistics; non-developer app building is covered in vibe coding statistics.
What Will AI Coding Assistants Look Like in 2027?
Axis Intelligence Research’s base case: 2027 is the first full year in which agents, not humans, start most pull requests on GitHub. Four data series point there.
- Agent pull requests. Octoverse shows the agent-started share rising from 17% (March) to more than 40% (September 2026): 3.83 percentage points a month. At that linear pace, the share crosses 50% about 2.6 months after September — around mid-December 2026. Because 40% is a floor, the crossover comes no later than that if the pace holds. (Axis Intelligence Research estimate; shares tend to slow as they approach saturation.)
- The 2028 forecast has already been passed. Gartner forecast in April 2024 that 75% of enterprise software engineers would use AI code assistants by 2028, up from under 10% in early 2023 (Gartner). JetBrains measured 90% weekly agent use among professional developers in mid-2026 — 15 points above the 2028 target, two years early. The populations differ, but the direction is not in doubt.
- The leadership gap is still widening. The Forward ICR™ (stated intent for the next six months) is 2.11, above every measured-usage reading. Unless intent reverses, the first 2027 survey waves should show a wider Claude Code lead over Copilot than 2026 did, with Codex the likeliest tool to close on second place (17.6% intent against Copilot’s 19.6%).
- Multi-model by default. 61% of Copilot users who picked a model used more than one in September 2026, up from 50% four months earlier, and the number of models used on paid plans rose from 29 in May to 47 in September (GitHub Octoverse 2026).
What would reverse the base case: a security incident traced to an unreviewed agent merge at a large enterprise, or a sharp rise in agent-PR rejection rates. Neither shows in the September data.
Related tracking: AI agents statistics.
Methodology
Collection. Axis Intelligence Research fetched every figure on this page from its producer on November 5, 2026: developer surveys (Stack Overflow 2026 and 2025; JetBrains Developer Ecosystem 2026 and AI Pulse January 2026), platform telemetry (GitHub Octoverse 2026), filings (Anthropic Form S-1; SpaceX Q3 2026 results furnished on Form 8-K), earnings calls (Microsoft FY26 Q2–Q4, FY27 Q1), company announcements (Anthropic, OpenAI), vendor pricing pages, controlled trials (METR) and security benchmarks (Veracode). Each figure is labeled filed, company-reported, survey or Axis calculation. No figure comes from an aggregator or a third-party statistics site.
Formulas.
- Incumbent Crossover Ratio (ICR™) = usage share of the leading agent-first challenger ÷ GitHub Copilot usage share, both from the same survey wave. Readings are never mixed across surveys. JetBrains values for April–June and September 2025 were read from the labeled points of JetBrains’ published charts.
- Forward ICR™ = the same ratio computed on Stack Overflow’s “want to use in the next six months” counts (5,073 ÷ 2,404).
- Intent-to-usage ratio = respondents who want to use a tool in the next six months ÷ respondents who used it in the past year, same Stack Overflow question (n=12,255).
- Power Tier Multiple (PTM) = highest self-serve individual monthly price ÷ lowest paid individual monthly price, monthly billing, as listed on November 5, 2026. For ChatGPT, Pro is priced “from $100” with higher usage levels unpriced on the page, so its PTM is a floor.
- Cursor annualized revenue = $608M × 365 ÷ 48.
- Switching rate = 100 − 33.7 (did not switch) − 0.6 (not applicable).
- Agent-PR crossover = (50 − 40) ÷ ((40 − 17) ÷ 6) months after September 2026.
- Awareness-to-work conversion = work use ÷ awareness, same JetBrains wave.
All arithmetic was run twice (floating point and exact fractions) before publication.
Scope. Survey shares come from self-selected (Stack Overflow) or quota-weighted (JetBrains) samples and are not interchangeable; this page compares them only through within-survey ratios. Stack Overflow’s 2025 tool question used a narrower base (AI agent users) than its 2026 question. Company user counts follow each company’s own definition. The Cursor annualization assumes revenue in the closing window is representative of a full year.
About This Dataset
The dataset behind this page — ai-coding-assistant-statistics.csv — holds 255 rows: every figure in the article with its value, unit, as-of date, source organization, document, URL, retrieval date, primary-source flag and, for Axis calculations, the formula. It covers survey waves from 2025 to October 2026, GitHub telemetry to September 30, 2026, filings to October 26, 2026, and prices verified November 5, 2026.
- License: CC BY 4.0. Free to reuse with attribution.
- Attribution line: Axis Intelligence Research, AI Coding Assistant Statistics 2026, 2026, axis-intelligence.com/ai-coding-assistant-statistics/
How to Cite This Research
APA: Axis Intelligence Research, & Mitchell, S. (2026, November 5). AI coding assistant statistics 2026: Agents, market leaders, revenue and the 2027 outlook. Axis Intelligence. https://axis-intelligence.com/ai-coding-assistant-statistics/
MLA: Axis Intelligence Research, and Sarah Mitchell. “AI Coding Assistant Statistics 2026: Agents, Market Leaders, Revenue and the 2027 Outlook.” Axis Intelligence, 5 Nov. 2026, axis-intelligence.com/ai-coding-assistant-statistics/.
Chicago: Axis Intelligence Research, and Sarah Mitchell. “AI Coding Assistant Statistics 2026: Agents, Market Leaders, Revenue and the 2027 Outlook.” Axis Intelligence, November 5, 2026. https://axis-intelligence.com/ai-coding-assistant-statistics/.
Frequently Asked Questions
Has Claude Code overtaken GitHub Copilot?
Yes, in work usage. JetBrains’ May–July 2026 survey puts Claude Code at 39% of professional developers versus 21% for GitHub Copilot, and Stack Overflow’s 2026 survey shows 65.5% versus 58.7% past-year use. Copilot still has the larger installed base, with 50 million users reported by Microsoft in July 2026.
Which AI coding assistant do developers want to use in 2027?
Claude Code, by a wide margin. In Stack Overflow’s 2026 survey, 41.4% of developers want to use Claude Code in the next six months, against 19.6% for GitHub Copilot and 17.6% for OpenAI Codex. Axis Intelligence Research’s Forward ICR™ puts Claude Code’s intent lead over Copilot at 2.11x.
What share of GitHub pull requests are started by AI agents?
More than 40% in September 2026, up from 17% in March, according to GitHub’s Octoverse 2026. That was 110 million agent-started pull requests in a single month. At the March–September pace, Axis Intelligence Research estimates the share passed 50% around mid-December 2026.
How much revenue does Cursor make?
SpaceX reported that Cursor contributed $608 million of revenue between August 14 and September 30, 2026, after acquiring it at a $60 billion implied equity value. Axis Intelligence Research annualizes that to about $4.62 billion a year. Cursor has not published a run-rate since the acquisition.
How much revenue does Claude Code generate?
Anthropic reported Claude Code run-rate revenue of over $2.5 billion on February 12, 2026, at least 17.9% of the company’s $14 billion run-rate at the time. Anthropic’s October 2026 S-1 does not break out a newer Claude Code revenue figure.
Is GitHub Copilot still priced per seat in 2026?
Only as a floor. Since June 1, 2026, Copilot Business ($19) and Enterprise ($39) per-user prices include an equal dollar amount of GitHub AI Credits, and agent usage beyond that is billed at model rates. Individual plans run from Pro at $10 to Max at $100 a month.
Why do developers switch AI coding tools so often?
Mostly for output quality. In Stack Overflow’s 2026 survey, 25.1% switched for better output and 9.7% for price-related reasons; Axis Intelligence Research calculates that 65.7% of developers switched tools in the past year.
Do AI coding agents make experienced developers faster?
The evidence is leaning positive but unproven. METR’s February 2026 update estimated an 18% speedup for returning developers, after a 19% slowdown in early 2025, but its confidence intervals include zero. METR also notes many developers now refuse to work without AI, which makes clean trials harder.
Is code written by AI coding assistants safe to ship?
Not without review. Veracode’s October 2026 tests found 62% of AI-generated samples passed security checks across 18 models, so 38% did not. The best models reached 73%.
How many developers will use AI coding assistants in 2027?
Professional adoption is already near saturation: JetBrains measured 90% weekly agent use in mid-2026, above Gartner’s 75%-by-2028 forecast. The 2027 question is depth, not reach — Axis Intelligence Research expects 2027 to be the first full year in which agents start most GitHub pull requests.