AI Coding Agent Pricing Compared Side by Side

Entry plans for the major coding agents cost $8 to $20 a month: Cursor Plus and Codex Go at $8, Copilot Pro at $10, Claude Pro at $17 on annual billing, Kiro Pro at $20. Heavy tiers sit at $100 to $200. The bigger difference is the meter: usage limits, credits or tokens. Prices checked 2026-10-31.

Prices checked 2026-10-31

Every figure here is from the vendor's own pricing page or docs, fetched on the check date, and each tool has a full page on this site.

Antigravity's pricing page lists a $0 plan, Google AI Pro and Google AI Ultra individual plans with a flexible AI credit pool, and Business plans starting at $30 per seat per month or a pay-as-you-go option with a $0 seat fee. Our fetch did not show Pro and Ultra prices. See Antigravity for the tool.

What each price is metering

The prices above are not comparable on their own. They buy different things.

Two patterns show up. Claude Code and Codex both draw on a pool shared with a chat product, so everything you do in chat counts against the coding allowance. Cursor, Copilot and Kiro run their own credit or pool systems, though Cursor shares its pool with Grok.

Entry plans

If you want Claude models in a flat plan, Claude Pro is the cheapest official route. If you want a low bill and are happy with Cursor's own models, Plus is $8. If you want broad model choice including Claude, Gemini, GPT, Grok and Kimi inside an IDE for a flat $10, Copilot Pro lists the widest set of models in its docs, though credit use depends on the model.

Heavy tiers

The allowances are expressed in different units, so you cannot read across. A Copilot credit is $0.01, so Copilot Max's 20,000 credits equal $200 of metered usage at Copilot's rates for a $100 price. A Kiro credit in the plan is $0.02, so Pro Max's 5,000 credits equal $100. Claude and Cursor describe their allowances as multipliers of a smaller plan. Codex describes message ranges. We cannot convert these to one number and neither can the vendors, so we do not try.

How overage works

This is the clearest place the tools differ, and the least visible on the pricing pages.

Kiro's overage is twice its in-plan price per credit. Copilot's is 1 cent against a base-credit price of 1 cent, with flex credits on top, so overage there is not marked up against the base. Claude's usage credits bill at the same API rates as any API user. Cursor's third-party models always bill at API list price, so there is no "included" tier for Claude or GPT on Cursor at all.

Team plans

Enterprise tiers in each case add audit logs, SCIM and data controls, and are priced by quote or by seat plus usage: Claude Enterprise is $20 per seat annually with usage at API rates and a 20-seat minimum, and Copilot Enterprise is $39 per granted seat. Copilot also offers "Copilot access", $0 for GitHub Enterprise Cloud users with fully metered use, or $10 for Standalone.

Model access

See Claude models, GPT models and Gemini models for the models behind these tools. For the raw token prices behind them, see LLM API prices.

A rough way to choose

Without usage tests we cannot say which tool gives the most work per dollar. We can say what is knowable from published pages.

What changed this year

Most of these plans moved recently, which is a reason to treat last year's comparison with suspicion.

• Cursor now sells through The X Subscription, with Free, Plus, Premium, Super and Ultra. Pro ($20), Pro+ ($60) and the old Teams seats are legacy plans that keep working.
• GitHub Copilot moved to usage-based billing with AI credits at $0.01, a base allowance matching the plan price, and a flex allotment that can change.
• Claude renamed Max 20x to Ultra at the same $200, and moved Fable on Pro to usage credits after a promotion ended on 19 July 2026.
• Codex is priced as part of ChatGPT, with Go at $8 and Pro tiers at $100, $200 and $500, and GPT-5.5 retiring from ChatGPT sign-in on 14 October 2026.
• Kiro sells credit-based plans from Free to Power, with a $20 first-upgrade bonus.
• Google moved individual Gemini CLI users to Antigravity CLI on 2026-06-18, and Gemini CLI is now Code Assist enterprise only.

Lock-in on price looks low, since these are monthly plans (annual billing is the exception on Claude Pro and Cursor). Project files such as rules and MCP configs may carry over between tools, but check the instruction file each tool reads before you switch. Codex and Claude Code both read an AGENTS.md or CLAUDE.md respectively.

Free options

Free tiers differ in what they restrict, and none says how long they last in practice. Use them to check that the workflow fits before paying.

Our take

Price pages are easy to compare and hard to use. Entry prices cluster between $8 and $20, and heavy tiers between $100 and $200, which makes tools look interchangeable. They are not. The meter differs, the overage rule differs, and the model access differs.

Three questions settle most decisions. Do you need Claude, GPT or both? If Claude, a Claude plan has the flat-price advantage, and Cursor or Copilot add model choice. Do you want a fixed bill or are you happy with a variable one? Flat plans cap exposure, and credit systems let you grow. And do you work alone or in a team? Seat prices cluster around $19 to $30.

We did not run these tools against each other, so we make no claim about quality or speed. Pick two, use each for a week on the same repository, and check the dashboards. For tool-level overviews see AI coding agents, Cursor alternatives and open-source AI agents.

Checklist

• Decide whether you need a particular model vendor, and shortlist tools that carry it.
• Find the overage rule before the headline price.
• Set a spend cap or budget on day one.
• Use free tiers for a week before paying annually.
• Re-check prices monthly. Several of these plans changed this year.

Reading the numbers without fooling yourself

Three habits keep a price comparison honest. First, convert everything to a monthly figure you would really pay. Claude Pro at $17 assumes $200 up front, Cursor's annual tiers assume a year's commitment, and Kiro's first month is reduced by the $20 sign-up credit. Second, ask what the headline price excludes. Cursor Plus excludes every third-party model. Claude's monthly API credits on Max and Ultra exclude Claude Code. Copilot's flex credits can change. Third, ask who bills the overflow. Anthropic bills usage credits at API rates, OpenAI bills Codex credits, GitHub bills a cent per credit, and Kiro bills four cents.

Then do a one-week trial on the same repository with the same kind of task, and note what each dashboard reports. Claude Code shows plan usage bars through /usage, Codex shows a usage dashboard and /status, Kiro shows credits used and remaining updated every five minutes, Copilot shows allowance and spend, and Cursor shows the weekly pool and reset date in editor settings. A week of dashboards tells you more than any pricing page, because the vendors themselves say usage depends on the task, the model and the context.

Finally, plan for a change. Several of these vendors changed their structure within the year, so avoid prepaying annually until you have used a tool through a heavy week. Where annual billing exists, it saves about two months on Cursor and about $40 on Claude Pro, which is real money but not worth being locked into a tool that does not fit.