Claude Code vs Codex CLI vs Cursor: What Each Actually Costs a Solo Developer per Month (2026)

Every few weeks someone asks me which AI coding tool is "cheapest." The honest answer is that the sticker price tells you very little. All three of the big options now meter you in some way: rolling windows, weekly caps, credit pools, or straight per-token billing. What you actually pay depends on how you work.

So I went through the official pricing pages for Claude Code, Codex CLI and Cursor (as of late October 2026) and wrote down what each one costs a solo developer, how the limits really work, and which plan I'd pick for three different usage patterns. Prices change constantly in this space, so treat this as a snapshot and check the current pricing page before you buy.

The big change most comparisons miss: Cursor's new plans

If you last looked at Cursor a month ago, the plan names you remember are gone for new customers. Since October 12, 2026, new Cursor subscribers buy it as part of The X Subscription, a combined plan for Grok, Cursor and X:

• Free: $0, limited Agent and Chat
• Plus: $8/month, Cursor's own models only (Grok and Composer), fixed settings
• Premium: $30/month, 4x Plus usage
• Super: $100/month, 15x Plus usage
• Ultra: $200/month, 50x Plus usage

Annual billing is $80, $300, $1,000 and $2,000 a year respectively.

The part that matters for cost: third-party models like Claude, GPT and Gemini are not in any plan's included pool. On Premium and above you can use them, but you pay the model's API list price from prepaid credit, starting with the first request. Plus and Free can't use them at all. The included pool covers Cursor's own models and refreshes weekly.

The old Pro ($20), Pro+ ($60) and Ultra plans still exist, but only for people who already had them. If you're on one, Cursor's docs say switching is optional and permanent, so read the legacy plans page before you click anything.

Claude Code

Claude Code comes with any paid Claude plan:

• Pro: $20/month billed monthly, or $17/month on annual ($200 up front)
• Max: from $100/month, with a choice of 5x or 20x Pro's usage. Max also includes monthly API credits ($100 to $200 depending on tier) for building on the Claude Platform. Check the pricing page for the exact price of the 20x tier.

How the limits work: usage resets on a rolling five-hour window, and paid plans add weekly limits on top. Chat on web, desktop and mobile shares the same pool as Claude Code, so a long research chat eats into your coding budget. There's no fixed message count; it depends on the model, context size and features. When you hit a limit you can wait, upgrade, or turn on usage credits billed at standard API rates.

If you'd rather skip subscriptions, Claude Code also works with pay-as-you-go API credits. Current list prices per million tokens on the Claude pricing page include Sonnet 5.5 at $2 input / $10 output and Opus 5.5 at $4 input / $20 output, with much cheaper cache reads.

Codex CLI

Codex is included with ChatGPT plans:

• Free: $0, and Go: $8/month, both limited to lighter models in the desktop app
• Plus: $20/month, with Codex on the web, CLI, IDE extension and iOS
• Pro: $100, $200 or $500/month tiers

On Plus, OpenAI publishes rough estimates of local messages per five-hour window: 15 to 160 for GPT-6.1 Sol, and 350 to 3,000 for the small GPT-6 Luna model. Weekly limits may also apply. The page says Pro plans currently have no five-hour limit. Cloud tasks use more of your allowance than local ones, and Fast mode burns included usage at 2.5x. You can check what's left mid-session with /status in the CLI.

When you run out, Plus and Pro users can buy extra credits, or run Codex with an API key. With a key you pay API prices: GPT-6.1 Sol is $2 input / $10 output per million tokens, with cached input at $0.10. API-key mode skips the cloud features like GitHub code review and Slack.

Side-by-side

	Claude Code	Codex CLI	Cursor (new customers)
Cheapest paid entry	Pro $20/mo ($17 annual)	Go $8/mo (limited); Plus $20/mo for full Codex	Plus $8/mo (own models only)
Mid tier	Max 5x, from $100/mo	Pro $100/mo	Premium $30/mo, Super $100/mo
Top tier	Max 20x (check page)	Pro $200 or $500/mo	Ultra $200/mo
Limit style	Rolling 5-hour window + weekly cap	5-hour window on Plus, weekly limits may apply	Weekly pool shared with Grok
Claude/GPT models	Claude only, included	OpenAI only, included	Pay API list price on Premium+
Overflow option	Usage credits at API rates	Buy credits or use API key	Top-offs, or prepaid credit for third-party models
Pure pay-per-token	Yes, via API/Console	Yes, via API key	Not really; it's a plan plus credit

Three usage profiles

1. The light user: a few sessions a week

You use an agent to scaffold a feature, write tests, or untangle a bug two or three evenings a week.

My pick: Claude Pro ($17 to $20) or ChatGPT Plus ($20). Either one should cover this comfortably, and you get the chat app too. Pick based on which model you like more for your stack.

If you want Cursor's editor experience and you're fine with Cursor's own models, Plus at $8 is the cheapest paid option here. Just know you're not getting Claude or GPT on that tier.

A pure API setup can also be very cheap at this level, but it has no ceiling. One runaway agent loop on a big repo can cost more than a month of Pro.

2. The daily user: an agent open most of the workday

You lean on the agent for real work every day, several multi-step tasks, some refactors, maybe a background task while you review.

This is where the five-hour windows start to bite on the $20 plans. You'll hit them on heavy days and spend some afternoons waiting.

My pick: start on the $20 plan of your preferred tool for two weeks and watch how often you hit the wall. If it's more than once or twice a week, the jump to Claude Max 5x or ChatGPT Pro at $100 usually pays for itself in time saved.

On Cursor, Premium at $30 gives you 4x Plus usage on Cursor's models. If your workflow depends on Claude or GPT inside Cursor, budget for API-price credit on top. That combined number is the one to compare against a $100 Claude or Codex plan.

3. The heavy agent user: parallel agents, long sessions

You run multiple agents at once, leave long tasks running, or use automations and cloud agents.

My pick: a top tier on whichever tool you use most, not two mid tiers. Claude Max 20x, ChatGPT Pro at $200, or Cursor Super/Ultra are all built for this. Codex Pro's lack of a five-hour limit (per the current page) is a real advantage if your usage is bursty.

What I would avoid is running heavy agent workloads on raw API billing without hard spend caps. At this volume, flat plans are almost always cheaper than per-token pricing for the same work.

Tips to cut spend

1. Shrink your context. Big CLAUDE.md or AGENTS.md files get re-sent constantly. OpenAI's own docs recommend nesting smaller AGENTS.md files instead of one giant one.
2. Turn off MCP servers you aren't using. Every connected server adds tool definitions to each request.
3. Use the small model for small jobs. Renames, boilerplate and test scaffolding rarely need the flagship model. On Codex Plus, the gap between Luna and Sol in the message estimates is huge.
4. Watch speed modes. Fast mode on Codex uses included usage at 2.5x. It's great when you need it and wasteful as a default.
5. Start fresh sessions. Long-running conversations keep dragging old context along. Clear or compact when you switch tasks.
6. Pay annually only after a month of data. Claude Pro annual and Cursor annual both save money, but only if you've confirmed the tier fits.
7. Set caps on anything per-token. API keys and Cursor's prepaid credit should have limits you chose deliberately.

Track your spend as you go

The single biggest thing that changed my bill was simply seeing my usage while I worked instead of at the end of the month. Some options:

• Built-in views. Claude shows plan usage in Settings > Usage, Codex has a usage dashboard plus /status in the CLI, and Cursor shows remaining usage in settings. They're accurate but you have to go look.
• ccusage (free, MIT). Run npx ccusage@latest and it reads the local logs from Claude Code, Codex and a long list of other agent CLIs, then prints daily, monthly, session and five-hour block reports with estimated costs. Everything stays on your machine. The costs are estimates, so they won't exactly match your bill.
• CodexBar (free, MIT, open source). A small macOS menu bar app that shows limits and reset times for Codex, Claude, Cursor and many other providers. If you want something always visible and free, start here.
• TokenBar. A Mac menu bar app that tracks Claude, Codex and other AI usage and cost in one place. It's built on the MIT-licensed CodexBar and costs $15 lifetime: tokenbar.site. Disclosure: I built TokenBar. If CodexBar or ccusage already covers what you need, use those.

Whichever you pick, give it two weeks. Most people I've talked to find their real usage is either much lower than the plan they're paying for, or concentrated in a few heavy days that a cheaper plan plus occasional credits would handle.

Bottom line

• Light use: a $20 plan (or Cursor Plus at $8 if you're happy with Cursor's own models).
• Daily use: start at $20, move to $100 only when the five-hour wall costs you real time.
• Heavy agent use: one top-tier plan on your main tool, never unlimited per-token billing.
• In Cursor, remember that Claude and GPT now cost API list price on top of any new plan.

And again, these numbers move fast. Before you commit, check the current pricing page for each tool.