What your AI coding plan really buys.

AI coding plans compared in one line: $20 a month buys a working allowance of one vendor’s coding agent, and the $100 and $200 tiers buy 5 times and 20 times that allowance. Anthropic, OpenAI and Google all use that ladder. Cursor charges $20, $60 and $200 for 1x, 3x and 20x its Pro agent limits, and GitHub Copilot sells monthly dollar credits instead.

What none of them publishes is how many tasks an allowance covers. This guide lines up what each plan includes, how its limits reset and what happens when you run out, using only the vendors’ own pages. It includes Cyborb, a desktop AI agent from Orbioom and our product, as one row marked (ours).

01What does a $20 AI coding plan get you?

A $20 plan gets you the vendor’s full coding agent with an allowance meant for part-time use, not all-day agent work. OpenAI describes ChatGPT Plus as enough to “power a few focused coding sessions each week,” and Anthropic pitches its higher Max tier at people who work with Claude all day.

Checked on September 28, 2026. Each plan name links to the page we read.

Only OpenAI and GitHub put hard numbers on the allowance: message ranges and dollar credits. Anthropic, Google, Cursor and Cyborb describe it as a multiple of a free plan or as “extended limits,” and even OpenAI calls its ranges estimates, not fixed limits.

02What do the $100 and $200 plans add?

Mostly more of the same allowance: the higher tiers keep the same agent and add usage, priority access and a few extras.

Checked on September 28, 2026.

Does a bigger plan give you more per dollar?

Only at the top. We divided each vendor’s stated usage multiple by its price multiple in a short Python script. The $100 tiers and Cursor’s $60 Pro+ buy exactly as much usage per dollar as the $20 plans. The $200 tiers at Anthropic, OpenAI, Google and Cursor buy twice as much, because they cost 10 times the $20 plan and promise 20 times the usage.

Two cautions. The multiples describe the short window (Anthropic says “per 5-hour session”) or agent limits (Cursor), and no vendor says weekly caps scale the same way. And the discount only counts if you use it: a $200 plan used at half pace costs the same per unit of work as a fully used $100 plan.

03How do the limits reset?

Claude, Codex and Google’s plans refill on a 5-hour clock and add a weekly cap (OpenAI says only that one may apply), while Cursor and GitHub Copilot hand out a monthly allowance.

Checked on September 28, 2026.

A 5-hour window forgives a heavy morning, but a weekly cap can still stop you on a Thursday. Our guide to AI usage limits explains both windows and how to make them last.

04What happens when you run out?

On most paid plans you can keep working by paying for the overflow; if you do not, you wait for the next reset.

• Claude: wait, upgrade, or turn on usage credits (Pro and Max), billed at standard API rates. Prepaid bundles save up to 30%. Claude Code may offer Console API credits, which you can decline.

• ChatGPT and Codex: the turn in progress can finish. Plus and Pro users can buy credits, valid for 12 months, and anyone can run extra local chats with an API key at API rates.

• Google: AI Pro and Ultra members can buy AI credits for Antigravity. A setting decides whether they are spent automatically or never.

• Cursor: on-demand usage at the same API rates, billed in arrears. Cursor says requests are never downgraded in quality or speed.

• GitHub Copilot: wait for next month, set a dollar budget (credits cost $0.01 each) or switch to a cheaper model. The Free plan cannot buy more.

• Cyborb (ours): wait for the window to reset, or move up a plan. Its pricing page lists no pay-as-you-go top-up.

Overflow billed at API rates is easier to judge once you know what a million tokens costs. Our LLM API pricing comparison has the current rates.

05Is there a free AI coding agent plan?

Yes. GitHub Copilot, Cursor, Codex, Google Antigravity, Kiro, Devin and Cyborb all have free tiers, but Claude’s free plan does not include Claude Code.

• GitHub Copilot Free: 2,000 completions a month and limited chat and agent use. Its FAQ puts chat at 50 requests.

• Cursor Hobby: limited Agent requests, no credit card needed.

• Codex on ChatGPT Free or Go ($8): GPT-6 Luna in the desktop app, subject to rollout.

• Google Antigravity: free, with a weekly quota across Gemini models, Claude Sonnet and Opus 4.6, and gpt-oss-120b.

• Kiro Free: 50 credits a month.

• Devin Free: a “light quota,” with limited model availability.

• Cyborb Free (ours): the Air model, metered usage that resets on its own, and 3 published sites.

06Which AI coding subscription gives the most use for $20?

Nobody can say exactly, because each vendor measures usage differently and most do not publish the size of the allowance.

What you can compare is cost per task on your own work. Run the same kind of task on two plans for a week and watch the meters. They live in Settings > Usage in Claude, /status in the Codex CLI, Cursor’s usage dashboard and your Copilot settings, which alert you at 75%, 90% and 100% of a budget.

07Claude Pro or Max for Claude Code?

Start on Pro and move to Max 5x only if you consistently hit your limits, which is also Anthropic’s own advice. Pro includes Claude Code, but chat and coding share one pool, so a long morning chat leaves less for the afternoon. If you run out only now and then, usage credits at API rates may cost less than the $80 step up to Max.

08Is ChatGPT Plus enough for Codex?

For a few focused sessions a week, yes, since that is how OpenAI describes Plus. Its estimates give Plus 15 to 150 GPT-6 Sol messages per 5 hours, against 70 to 700 on the $100 Pro tier. The efficient GPT-6 Luna model stretches that to 350 to 3,000 messages on Plus. If you still run out, note that as of late September 2026 only the $100 Pro tier is open to new subscribers.

09Cursor Pro, Pro+ or Ultra?

Cursor recommends Pro+ for daily agent users and Ultra for power users. Its docs add a useful yardstick: daily Agent users typically spend $60 to $100 a month in total usage, and power users running several agents often spend $200 or more. Tab completions are unlimited on Pro, Pro+ and Ultra, so light users rarely need more than Pro.

10Other coding plans worth a look

Two more agents appear in most comparisons, and both sell the same kind of ladder. For a wider look at the tools themselves, see the best AI coding agents of 2026.

Checked on September 28, 2026.

11FAQ

Read next: Claude Code vs Codex vs Cursor to pick the tool, how to choose an AI model to spend your allowance well, and every AI subscription price on one page.