Claude Code vs Codex vs Cursor Pricing and Usage Limits

Short answer: for most solo builders the cheapest serious setup is Claude Pro ($20/month, or $17 billed yearly) for Claude Code, or ChatGPT Plus ($20) for Codex. Cursor Premium costs $30 and its subscription no longer covers Claude or GPT models at all: those are billed on top at API list price. If you code with an agent all day, Claude Max ($100 or $200) or ChatGPT Pro (from $100) is usually cheaper than paying per token with an API key.

The rest of this guide shows exactly what each plan includes, how usage limits work, what an API key costs instead, and which tool fits which kind of builder. All prices were checked on 28 October 2026 and exclude tax. Prices in this category change often, so confirm on the vendor page before you buy.

The three tools in one sentence each

• Claude Code is Anthropic’s terminal and IDE coding agent. It is bundled into every paid Claude plan and draws from the same usage pool as Claude chat (claude.com/pricing).
• Codex is OpenAI’s coding agent (CLI, IDE extension, desktop app and cloud). It is bundled into ChatGPT plans and shares usage with ChatGPT Work (Codex pricing).
• Cursor is an AI code editor. Since it became part of The X Subscription, one plan covers Cursor, Grok and X, and Cursor and Grok share a weekly usage pool (Cursor help).

That last change matters. Many comparison pages still describe Cursor’s older Pro and Pro+ plans with included Claude and GPT usage. Cursor’s docs now call those legacy plans: they keep working for existing customers, but new customers start on The X Subscription (Cursor docs).

Price table: individual plans (as of October 2026)

Checked 28 October 2026. USD, excluding tax.

Sources: Claude pricing, Max plan help article, Codex pricing, Cursor pricing, Cursor docs. Cursor annual billing is $80/yr (Plus), $300/yr (Premium), $1,000/yr (Super) and $2,000/yr (Ultra).

Team plans

• Claude Team: Standard seat $20/seat/month billed yearly ($25 monthly); Premium seat $100 yearly ($125 monthly) with 5x Standard usage. Teams of 2 to 150, with up to $500/month in pooled API credits.
• ChatGPT Business: $20/user/month billed yearly, $25 monthly, minimum 2 users.
• Cursor Teams: $30, $100 or $200 per seat, matching Premium, Super and Ultra. Teams also pay a Cursor Token Rate of $0.25 per million tokens on eligible third-party models.

How usage limits actually work

None of the three sells “unlimited” agent use. Each plan buys a share of compute that resets on a schedule.

Claude Code. Every Claude plan has limits that reset on a rolling five-hour window, and paid plans add weekly limits. Claude Code, chat, desktop and mobile all draw from the same pool. Anthropic gives no fixed message count, because usage depends on conversation length, model and features. When you hit a limit you can wait, upgrade, or turn on usage credits billed at standard API rates (claude.com/pricing).

Codex. OpenAI publishes rough ranges instead of hard numbers. For ChatGPT Plus, the estimate is 15 to 160 local messages per five hours on GPT-6.1 Sol, 5 to 45 on GPT-6 Astra, and 350 to 3,000 on the small GPT-6 Luna. Weekly limits may also apply. Pro plans currently have no five-hour limit. Plus and Pro users can buy credits to keep going (Codex pricing).

Cursor. Paid plans include one usage pool for Cursor’s own models, shared across the editor, CLI, cloud agents and the Grok apps. It refreshes weekly. Third-party models (Claude, GPT, Gemini) are outside the pool entirely and bill at API list price from prepaid credit (Cursor docs).

The practical upshot: if you want Claude Opus inside Cursor, you pay Cursor’s subscription and Opus API rates. If you want Opus in Claude Code, a Claude plan covers it until you hit the session or weekly cap.

API key vs subscription billing

Both Claude Code and Codex can run on an API key instead of a plan.

• Claude Code with a Console account bills every token at API rates. Current rates per million tokens: Opus 5.5 $4 input / $20 output, Sonnet 5.5 $2 / $10, Haiku 5.5 $0.10 / $0.50 (prompts up to 100K tokens). Prompt-cache reads are much cheaper, for example $0.20 per million on Opus 5.5 (claude.com/pricing).
• Codex with an API key works in the CLI, SDK and IDE extension, but you lose cloud features such as GitHub code review and the Slack integration. Usage bills at OpenAI API pricing (Codex pricing).
• Cursor individual plans do not take your own API keys on Plus. On Premium and above, third-party models are already billed at API list price; individual plans do not pay the extra Cursor Token Rate (Cursor docs).

OpenAI warns not to use API token prices to estimate how much a subscription includes. The two are metered separately.

What it costs per task: cite the benchmark, not a guess

We have not run our own head-to-head tests. For capability and cost per task, the most useful public data is the Artificial Analysis Coding Agent Index, which runs agents on three benchmarks (FrontierCode v1.1, Terminal-Bench 4.0 and SWE-Atlas-QnA-AA) and reports pay-per-token API cost per task.

From the Claude Code vs Codex comparison, checked 28 October 2026:

Three things to read from this. Claude Code scored highest but burned far more tokens per task. Codex on GPT-6.1 Sol came within 8 index points for about one-eighth of the cost. And these are hard benchmark tasks, so your everyday fixes will usually cost less. Artificial Analysis itself notes the cost figure is API pricing, not plan pricing. We could not confirm a published Cursor result on the same index, so we do not compare Cursor on capability here.

Worked example: a solo founder shipping 20 large tasks a month

Assume you hand the agent 20 substantial tasks a month, each similar in size to a benchmark task. Using the API cost per task above:

• Claude Code on Opus 5.5 via API key: 20 × $11.42 = $228.40/month.
• Claude Max 5x: $100/month flat, as long as you stay within session and weekly limits. That is $128.40 less than the API route.
• Codex on GPT-6.1 Sol via API key: 20 × $1.50 = $30/month. Here ChatGPT Plus at $20 is only $10 cheaper, so the API key is a reasonable choice if your usage is spiky.
• Cursor Premium using Claude Opus 5.5: $30 subscription + Opus at the same API list price ($4 / $20 per million tokens per Cursor’s docs), so roughly $30 + $228.40 = about $258/month for the same Opus workload.

Now scale down to 5 tasks a month on Opus: 5 × $11.42 = $57.10 via API, which is cheaper than Max 5x but more than Pro at $20. Pro is the better buy if its limits are enough for you.

The rule of thumb: the more you use a frontier model, the more a flat plan saves you. The cheaper the model, the less the flat plan matters.

Which to pick, by use case

Who should not switch

If you already pay for a legacy Cursor Pro or Pro+ plan with included Claude usage, Cursor says your plan keeps working unless you change it, and switching to The X Subscription is permanent (Cursor help). Do the arithmetic before you move.

If you pay from Canada or Australia, budget for sales tax on top of these USD prices (GST/HST in Canada, 10% GST in Australia) plus any card foreign exchange fee. We break that down, with CAD and AUD conversions, in Cursor pricing explained. If your team mainly needs AI inside email, docs and meetings rather than a coding agent, compare Google Workspace Gemini pricing, where Gemini is now bundled into the plans. For app builders that write the code for you instead, see Lovable vs Replit vs Bolt pricing, and for workflow tools see n8n vs Make vs Zapier pricing. More guides live in the AI tools hub.

Bottom line

Start on the $20 tier of whichever model family you prefer, watch your usage page for two weeks, and only upgrade when you hit a limit more than once a week. Use API keys for automation, not for your daily driver. And treat Cursor’s new pricing as an editor subscription plus API bills, not a bundle.

If you call the models through their APIs instead, price your usage with the free LLM API Cost Calculator.