Claude Code vs Codex vs Cursor vs Antigravity vs Copilot: the short answer
Cursor's cheapest paid plan now costs $8 a month and cannot run a single model from Anthropic, OpenAI or Google. It is the sharpest of the changes since our September comparison, and it arrives as searchers add a fifth name to the list: GitHub Copilot.
The short answer for early November 2026: Claude Code is the agent for long, tightly coupled changes, because every paid plan starts on Claude Opus 5.5, the highest-scoring model on Artificial Analysis's independent index that anyone can buy. Codex is for people who want published limits and a wide price range, and it now defaults to GPT-6.1 Sol. Cursor is the editor for people who still write much of the code themselves, and it is cheapest on its own Grok models. Antigravity is free to start and built for supervising several agents, and paying subscribers now get Gemini 4 Argon and Claude Opus 5.5 in it. GitHub Copilot puts models from all of those labs behind one subscription that starts at $10, inside almost any editor, and it is the natural home for teams whose work already lives on GitHub.
None of the five wins everywhere. What separates them is where each one wants you to work and which model it hands you before you change a setting.
| Tool | Where it runs | Default and notable models | Individual plans, monthly | Published limits |
|---|---|---|---|---|
| Claude Code | Terminal; VS Code (also Cursor) and JetBrains extensions; desktop app; web and mobile cloud sessions | Opus 5.5 by default; Sonnet 5.5 and Haiku 5.5; Fable 5.1 on credits (Pro) or up to half the weekly limit (Max and Ultra) | Pro $20; Max $100; Ultra $200; Team seats $25 or $125 | Multiples only: Max is 5x Pro per five hours and per week; Ultra is 10x Pro per week, with no five-hour limit once verified |
| Codex | Open-source CLI; IDE extension; ChatGPT desktop app; Codex Cloud on web and mobile | GPT-6.1 Sol by default in the CLI; GPT-6 Astra, GPT-6.1 Terra, GPT-6 Luna | Free and Go ($8) with Luna; Plus $20; Pro $100, $200 or $500; Business $25 a seat | Plus, per five hours: 15 to 160 GPT-6.1 Sol, 35 to 285 Terra, 5 to 45 Astra, 350 to 3,000 Luna messages; Pro has no five-hour limit |
| Cursor | AI editor; CLI; cloud agents; Projects (beta) | Auto; Grok 4.7, 4.6, 4.5 and Composer 2.5 in the included pool; Claude 5.5 models, Fable, GPT-5.6 and Gemini 4 Argon at API prices | Free; Plus $8 (Cursor models only); Premium $30; Super $100; Ultra $200 | Weekly pool: Premium 4x, Super 15x, Ultra 50x Plus |
| Antigravity | Antigravity 2.0 desktop app; IDE; CLI; SDK | Gemini 3.8 Flash on the free plan; Gemini 4 Argon, Claude Opus 5.5 and Sonnet 5.5 on Google AI Pro and up | Free; Google AI Pro $19.99; Ultra $99.99 or $199.99 | Weekly rate limits, no numbers; Ultra is 5x or 20x Pro |
| GitHub Copilot | VS Code, Visual Studio, JetBrains, Xcode, Eclipse; Copilot CLI; Copilot app; cloud agent on GitHub | Auto selection; Sonnet 5.5, Haiku 5.5, GPT-6.1 Terra, GPT-6 Luna, Gemini 3.8 Flash, Grok 4.7 on Pro; Opus 5.5, GPT-6.1 Sol, Astra, Gemini 4 Argon from Pro+ | Free; Pro $10; Pro+ $39; Max $100; Business $19 and Enterprise $39 a seat | AI credits at $0.01, spent at API rates: Pro 1,500, Pro+ 7,000, Max 20,000 a month |
Prices are in US dollars on monthly billing, as listed on 3 November 2026 by Anthropic, OpenAI, Cursor, Google and GitHub.
| Cursor Plus (Grok and Composer only) | $8 |
|---|---|
| ChatGPT Go (Codex on GPT-6 Luna only) | $8 |
| GitHub Copilot Pro | $10 |
| Google AI Pro (Antigravity) | $19.99 |
| Claude Pro (Claude Code) | $20 |
| ChatGPT Plus (Codex) | $20 |
What changed since September
Anthropic released Claude Sonnet 5.5 on 28 September at Sonnet 5's price of $2 and $10 per million tokens, saying it generates output more than 30% faster and costs up to 30% less per task. Claude Haiku 5.5 followed on 7 October at $0.10 and $0.50 for prompts up to 100,000 tokens, a tenth of Haiku 4.5's price.
OpenAI released GPT-6.1 Sol at DevDay on 29 September, and the Codex CLI made it the default model the same day. OpenAI also reopened its $200 Pro plan to new subscribers with a smaller allowance, added a $500 plan with its Ultrafast speed tier, and retired GPT-5.5 from Codex for ChatGPT sign-ins on 14 October. Our OpenAI DevDay 2026 recap covers the rest.
Google made Gemini 4 Argon generally available on 13 October. Its announcement offered the model in Antigravity to Google AI Ultra subscribers; Antigravity's documentation now lists it for Google AI Pro too, alongside Claude Opus 5.5 and Sonnet 5.5. GitHub added Haiku 5.5 to Copilot on 7 October and Gemini 4 Argon on 13 October.
Four more things have changed since this page was last updated on 22 October. OpenAI added GPT-6.1 Terra to Codex on 26 October for Plus subscribers and up, at $1 and $5 per million tokens on the API. Artificial Analysis moved its index to a new version on 27 October, which reshuffled the scores further down this page. Google announced Gemini 4 Carbon on 28 October, but gave access only to a subset of cyber defenders, with developers to follow "as soon as possible"; none of the five tools offers it yet. And Anthropic renamed its $200 Max 20x plan as Ultra, which gives ten times Pro's weekly usage and drops the five-hour limit for verified accounts.
Which model does Claude Code use now?
Claude Code still works best in a terminal, but it now runs in several places: the CLI, extensions for VS Code (which also install in Cursor) and JetBrains, a desktop app, and the web at claude.ai/code, where cloud sessions keep running after you close the laptop. Cloud scheduled tasks, which Anthropic used to call routines, run jobs on a timer or when something happens on GitHub.
Its default model is the strongest of the five. Claude Code's model documentation says Pro, Max, Ultra, Team, Enterprise and API accounts all start on Opus 5.5 at medium effort, and on Anthropic's API the sonnet and haiku aliases now resolve to Sonnet 5.5 and Haiku 5.5. Fable 5.1 costs more than Opus, but Anthropic's help center says Pro users reach it only through paid usage credits, and Max and Ultra users can spend up to half their weekly limits on it. Artificial Analysis currently scores Fable 5.1 below Opus 5.5.
Sonnet 5.5 is the reason to open the model menu. Anthropic puts it at 70.6% on Terminal-Bench 4.0, an agentic coding test, against 66.4% for Opus 5.5, at half the per-token price. Artificial Analysis also found it ahead, 64% to 60%, but measured it using about 193,000 output tokens per task on its index, the most it has recorded. A cheaper token does not guarantee a cheaper job. Haiku 5.5, at a fortieth of Opus's price, suits bulk edits that need little judgement.
The rest of Claude Code's edge is packaging: a CLAUDE.md file for standing instructions (it can read an AGENTS.md file instead), skills that turn a repeatable job into a command the whole team runs, hooks, and background agents that run sessions side by side. The weak spots are unchanged. Anthropic publishes multiples rather than message counts: Max gives five times Pro's usage every five hours and every week, and Ultra, the renamed Max 20x, gives ten times Pro's weekly usage with no five-hour limit once Anthropic has verified the account. Neither number tells you what Pro itself buys, so you learn your real limit by hitting it. The command sandbox runs on macOS, Linux and WSL2, not on native Windows. Claude Code fits long, tightly coupled changes, and teams that want the same job done the same way every time. Our guide to using Claude Code for SEO shows what that looks like on crawl, Search Console and Ads exports.
What does Codex run on after DevDay?
Codex comes with ChatGPT plans and runs in the same places as Claude Code: an open-source CLI, an IDE extension, the ChatGPT desktop app, and Codex Cloud, where each task gets its own workspace and keeps going while your computer sleeps. DevDay, according to OpenAI's recap, added voice control in the CLI, code review in the desktop app with automatic first-pass reviews in the cloud, and Codex Security Cloud, a research preview that scans connected GitHub repositories and watches new commits.
Its model range is the widest. OpenAI's API prices run from GPT-6 Astra at $10 and $50 per million tokens, through GPT-6.1 Sol at $2 and $10 and the new GPT-6.1 Terra at $1 and $5, to GPT-6 Luna at $0.10 and $0.50. OpenAI's model guide describes GPT-6.1 Sol as "near-Astra performance for complex work at a lower cost than Astra" and pitches Terra at drafting, charts and everyday app work rather than hard coding. The company says Sol matches Astra on DeepSWE v1.1, a test in real codebases, at about a fifth of the cost. That is a vendor result on vendor settings, but it answers September's complaint that matching Claude Code's default meant choosing Astra and its small allowance.
Codex is the only tool here that publishes message counts: on the $20 Plus plan, 15 to 160 GPT-6.1 Sol messages per five hours, 35 to 285 with Terra, 5 to 45 with Astra and 350 to 3,000 with Luna. Pro, now $100, $200 or $500, has no five-hour limit, though other limits may apply. OpenAI's Pro tiers page says only the $500 plan gets Ultrafast, which OpenAI says generates tokens up to eight times faster in Codex, and that the $200 plan is open to new subscribers again with a lower allowance. Earlier subscribers kept the old allowance until 29 October.
Codex also runs natively in PowerShell with its own Windows sandbox, so WSL is not required. It fits high volumes of small, well-specified tasks, Windows-heavy teams, and anyone who wants to plan against published limits. Our Claude Code vs Codex comparison goes through the two in detail.
What do Cursor's new plans buy?
Cursor is for reading and changing code yourself with AI close at hand: completions, inline edits and agents in an editor, plus a CLI, cloud agents and Projects, a beta in which a coordinator agent splits a large job among other agents. Since September its changelog has added remote control of local agents from Cursor's iOS app.
The plans are what changed. Cursor, which SpaceX acquired in August, is now sold as part of The X Subscription, which its pricing documentation describes as one subscription for Grok, Cursor and X. Plus costs $8 and covers only Cursor's own models (Grok 4.7, 4.6 and 4.5 and Composer 2.5), with Fast mode off and medium effort. Premium at $30, Super at $100 and Ultra at $200 carry four, fifteen and fifty times Plus's weekly usage, and only those tiers can use Claude, GPT or Gemini models, at API list prices from prepaid credit. The old Pro and Pro+ plans continue for existing customers only. Team seats cost $30, $100 or $200 a user, against $40 and $120 in September.
Grok 4.7 costs $2 and $6 per million tokens beyond the weekly pool, and Composer 2.5 costs $0.50 and $2.50. Artificial Analysis scores Grok 4.7 at 35 on its index against 58 for Opus 5.5, so the included pool buys a capable model rather than a frontier one. Cursor's model price list now includes Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1 and Gemini 4 Argon, but its OpenAI entries are still GPT-5.6 models, with no GPT-6 or GPT-6.1.
Cursor fits people who spend the day in an editor and want the cheapest capable model for routine agent work. If your work depends on Claude or GPT models, you pay API rates on top of the subscription, and for heavy Opus 5.5 use a plan sold by the model's maker is the simpler bill.
Is Antigravity better than Cursor now?
For work that splits into independent tasks you can review one at a time, and for trying agents without paying, yes. For a developer who wants completions while writing code, Cursor is still the better tool. The difference is who does the typing. Cursor assumes you are in the file. Antigravity 2.0, the desktop app Google released for macOS, Linux and Windows in May, has no editor at all: you hand tasks to agents, group them into projects and review what comes back. The original Antigravity IDE remains, and an Antigravity CLI replaced Gemini CLI.
The model list is what changed. Antigravity's model documentation lists Gemini 4 Argon for Google AI Pro, Ultra and Enterprise, and Claude Opus 5.5 and Sonnet 5.5 for Pro subscribers outside a trial, Ultra and Enterprise. In September its third-party models were a generation behind. A $19.99 Google AI Pro subscription now reaches the models Claude Code starts on, at limits Google does not publish. The free plan lists Gemini 3.8 Flash alongside Claude Sonnet and Opus 4.6 and gpt-oss-120b, which the documentation marked for removal on 2 November, so free users should plan on Gemini 3.8 Flash alone.
Google's pricing page describes the limits only as basic, more and even more, and its May subscription announcement put the two Ultra tiers at five and twenty times Pro. Cursor publishes multiples for every tier, which makes its bill easier to predict. Recent changelog entries add a /plan command and file and network sandboxing on Windows. Antigravity fits work that divides cleanly and teams already on Google Cloud. Our Cursor vs Antigravity comparison sets the two side by side in more detail.
Where does GitHub Copilot fit against the other four?
GitHub Copilot is the tool many developers already pay for, often through an employer, and it is now much more than autocomplete. Its plans page lists agent mode and the Copilot CLI on every plan, and a cloud agent and pull request reviews on paid ones. The cloud agent works in an ephemeral cloud environment and takes work from a chat or from an issue or pull request on GitHub. The Copilot app, a desktop client for macOS, Windows and Linux, runs parallel agent sessions in separate worktrees. And Copilot runs in more editors than the others, including Visual Studio, JetBrains IDEs, Xcode and Eclipse.
Its pricing works differently. Since 1 June every Copilot plan has come with GitHub AI Credits, spent on tokens at each model's published API rate; code completions do not use them. A credit is worth $0.01. Pro costs $10 and includes 1,500 credits a month, Pro+ costs $39 for 7,000, and Max costs $100 for 20,000, which is $200 of model usage at list price. Business is $19 a seat with 1,900 credits and Enterprise $39 with 3,900.
The model list is the broadest of the five, with one catch. GitHub's supported models table gives Pro Claude Sonnet 5.5 and Haiku 5.5, GPT-6.1 Terra, GPT-6 Luna, Gemini 3.8 Flash and Grok 4.7, among others; Claude Opus 5.5, GPT-6.1 Sol, GPT-6 Astra and Gemini 4 Argon start at Pro+. Free accounts get models only through automatic selection, and paid plans get a 10% discount on model costs when they let Copilot choose. Paid plans can also hand tasks to Anthropic's Claude and OpenAI's Codex agents from inside GitHub once they are enabled in the account's policies, a public preview billed through Copilot's credits and Actions minutes. Local sandboxing for agent commands has been generally available on Windows, macOS and Linux since 7 October.
That makes Copilot the cheapest way to try several labs' models on real work, the strongest choice for teams on Visual Studio, Xcode or Eclipse, and the natural home for teams whose tickets and reviews already live on GitHub. Its weakness is the other side of the credit model. Agent work on frontier models draws credits at full API rates, so a long Opus 5.5 session costs what it would on Anthropic's API, and the $15 of usage in Pro will not go far. Anthropic and OpenAI do not publish a dollar value for their subscription allowances, so the honest comparison is to run the same week of work on both and read the meters.
What the benchmarks say about the models underneath
The independent scores were reset at the end of October. Artificial Analysis moved to version 5 of its Intelligence Index on 27 October, adding a coding test graded on whether a repository maintainer would merge the agent's change and dropping three older evaluations. On its leaderboard the next day, Google's Gemini 4 Carbon took first place at 59, one point ahead of Claude Opus 5.5. GPT-6 Astra scores 57, GPT-6.1 Sol 55, Claude Sonnet 5.5 54, GPT-6.1 Terra 51, and Gemini 4 Argon and Claude Fable 5.1 50. The small and house models sit far lower: Haiku 5.5 at 36, Grok 4.7 at 35 and GPT-6 Luna at 28.
| Gemini 4 Carbon (restricted) | 59 |
|---|---|
| Claude Opus 5.5 | 58 |
| GPT-6 Astra | 57 |
| GPT-6.1 Sol | 55 |
| Claude Sonnet 5.5 | 54 |
| GPT-6.1 Terra | 51 |
| Gemini 4 Argon | 50 |
| Claude Fable 5.1 | 50 |
| Claude Haiku 5.5 | 36 |
| Grok 4.7 | 35 |
| GPT-6 Luna | 28 |
Those are best-case scores, and none of these tools ships at maximum effort. Claude Code runs Opus 5.5 at medium by default, where the same leaderboard gives it 48; GPT-6.1 Sol at medium scores 46. Moving either model from medium to maximum adds nine or ten points, against a gap of two or three between them, so the effort setting moves the score more than the choice between the two tools' default models does. The leaderboard also prices a run of its tasks at API rates: $2.45 per task for Opus 5.5 at medium and $0.36 for GPT-6.1 Sol at medium.
Agentic coding scores shift with whoever runs the test. Anthropic reports Sonnet 5.5 at 70.6% on Terminal-Bench 4.0 and Opus 5.5 at 66.4%. Artificial Analysis, on its own setup, got 64% and 60%, with GPT-6 Astra at about 60% and, in its Gemini 4 Argon review, Argon at 57%. The order survives the change of test bench; the margins shrink. Because token use per task varies so widely, cost per finished job is the number to watch, not price per token. For what the three frontier models cost on search and ads jobs, see our Gemini 4 Argon vs Claude Opus 5.5 vs GPT-6.1 Sol comparison.
Which AI coding tool for which kind of work
Take Harbour & Finch LLP, a sample Toronto law firm we use across this site (the firm and its numbers are invented): a 400-page Next.js site with a broken title template, product pages missing structured data, a migration that needs a redirect map, and a marketing team that wants a weekly script joining Search Console and Google Ads exports. Each job points at a different tool.
The title template and the structured data are one long change across many files, where each step depends on the last. That is Claude Code's strongest ground, on Opus 5.5 or, for most of the edits, Sonnet 5.5, with a person reviewing the diff before it ships. Codex on GPT-6.1 Sol is the closest alternative, and the better one if the team works on Windows.
The redirect map is hundreds of small, well-specified decisions: old URL, new URL, status code. A small model handles that at a fraction of a frontier model's cost, whether GPT-6 Luna in Codex, with up to 3,000 Plus messages per five hours, or Haiku 5.5 in Claude Code. Either way, a script should check that every target URL resolves before the map goes live.
Writing the reporting script is a well-defined issue you can assign to Copilot's cloud agent if the repository lives on GitHub; running it every Monday is a job for Claude Code's cloud scheduled tasks or a plain cron job. Anything in it that touches money should have code compute the figures and a model explain them, as our piece on Claude Opus 5.5 for SEO and Google Ads sets out. That division of labour, with code producing every figure and a person approving every change, is how AIO Copilot (no relation to GitHub's) handles SEO, AI search and Google Ads work on Claude and Gemini. It is priced per site; tell us how many you manage and we email the price.
The developer who maintains the site day to day, reading code and making small changes by hand, is Cursor's customer, on the $8 plan if Grok models are good enough for the work. An agency with many client repositories and a mix of editors gets the most from Copilot Business: one seat price, models from every major lab, and agents started from the issues where the work is already tracked. Antigravity is the cheapest way to find out whether supervising agents suits a team at all, with Opus 5.5 and Gemini 4 Argon a $19.99 subscription away.
Most developers who use these tools seriously settle on two: an editor for the day's work and an agent for the long jobs. Every entry tier here costs $20 or less, and Antigravity and Copilot both have free plans, so a month testing a pair on five real tasks from your own backlog is cheap. Count merged changes, review time and whether you hit a limit, then choose, and expect to choose again before long. In the twelve days between the last two versions of this page, one lab released a model, another announced one that few people can use yet, a plan was renamed and the main independent index was rebuilt.
Questions
What is the best AI coding tool in November 2026?
There is no single winner. Claude Code has the strongest default model, Claude Opus 5.5. Codex publishes its usage limits and defaults to GPT-6.1 Sol. Cursor suits people who work in an editor all day and is cheapest on its own Grok models. Antigravity is free to start and offers Gemini 4 Argon and Claude Opus 5.5 to paying subscribers. GitHub Copilot puts models from Anthropic, OpenAI, Google and xAI in one subscription from $10 a month.
Which is better, Claude Code or Codex?
For long, tightly coupled changes, Claude Code, because every paid plan starts on Opus 5.5. For many small, well-specified tasks, native Windows work or planning against published limits, Codex, which lists 15 to 160 GPT-6.1 Sol messages per five hours on the $20 Plus plan. Both run in a terminal, an IDE, a desktop app and the cloud, and Claude Code can read the AGENTS.md file that Codex uses.
What model does Codex use by default?
GPT-6.1 Sol. The Codex CLI made it the default model on 29 September 2026, the day OpenAI released it at DevDay, and OpenAI recommends it for complex coding. Its API price is $2 per million input tokens and $10 per million output tokens. GPT-5.5 was retired from Codex for ChatGPT sign-ins on 14 October.
Does Google Antigravity use Gemini 4 Argon?
Yes, on paid plans. Google made Gemini 4 Argon generally available on 13 October 2026, and Antigravity's model documentation lists it for Google AI Pro, Ultra and Enterprise. The free plan does not include it. Gemini 4 Carbon, which Google announced on 28 October, is not in Antigravity: Google is releasing it first to a small group of cyber defenders.
Is Google Antigravity free?
Yes, for individuals, with basic weekly rate limits. The free plan lists Gemini 3.8 Flash, plus Claude Sonnet and Opus 4.6 and gpt-oss-120b, which Google marked for removal on 2 November 2026. Google AI Pro ($19.99 a month) and Ultra ($99.99 or $199.99) raise the limits and add Gemini 4 Argon, Claude Opus 5.5 and Claude Sonnet 5.5.
How much does Cursor cost now?
Cursor's individual plans are Free, Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200, sold as part of The X Subscription. Plus covers only Cursor's own Grok and Composer models. Premium and above add Claude, GPT and Gemini models at API list prices, paid from prepaid credit. The older Pro and Pro+ plans continue for existing customers.
How much does GitHub Copilot cost, and which models does it include?
Copilot Free costs nothing, Pro $10 a month, Pro+ $39 and Max $100; Business is $19 and Enterprise $39 per seat. Each plan includes GitHub AI Credits worth $0.01 each, spent at each model's API rate. Pro includes Claude Sonnet 5.5 and Haiku 5.5, GPT-6.1 Terra, GPT-6 Luna, Gemini 3.8 Flash and Grok 4.7. Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon need Pro+ or above.
Can I use Claude Code and Codex inside GitHub Copilot?
Partly. Paid Copilot subscribers (Pro, Pro+, Max, Business and Enterprise) can assign tasks to Anthropic's Claude and OpenAI's Codex coding agents from GitHub once the agents are enabled in their policies, in public preview, billed through Copilot's AI credits and GitHub Actions minutes. Claude Code and Codex themselves also run in a terminal beside any editor, so many developers use one of them next to Copilot or Cursor.
New studies, by email
Our own studies, and what each new release changes for SEO, AEO and Google Ads.
