← the ledger · Which AI coding plan has the best limits?

Which AI coding plan has the best limits? (2026)

Sources: official pricing pages for Claude, Cursor, GitHub Copilot, OpenAI Codex, Gemini, Devin & Mistral · figures checked 2026-06-14 · prices in USD

If you code with an AI assistant, the limit is the product. A plan that runs out at lunchtime is worthless no matter how good the model is. So "which coding plan has the best limits?" should have a clean answer. It doesn't — and the reason is the finding worth knowing: almost no vendor publishes its coding limit as a number you can compare. Most say "more usage", "extended limits", or a bare multiplier. Below is every major coding plan, what it costs, and exactly what each one tells you (and refuses to tell you) about the limit.

The coding plans, side by side

PlanPriceWhat it says about the coding limit
GitHub Copilot Free$0"2,000 completions per month" · Copilot CLI included real number
GitHub Copilot Pro$10/mo"$15 monthly total credits" · unlimited code completion · sign-ups paused
GitHub Copilot Pro+$39/mo"$70 monthly total credits" · 4x+ usage vs Pro · sign-ups paused
GitHub Copilot Max$100/mo"$200 monthly total credits" · 2.9x+ usage vs Pro+ · sign-ups paused
Mistral Pro (Vibe)$14.99/mo"All-day coding in the CLI, IDE, and on web" · "subject to fair usage limits"
Claude Pro (Claude Code)$17/mo"Includes Claude Code and Claude Cowork" · "Usage limits apply" (no number)
Cursor Pro$20/mo"Extended limits on Agent" · included pool, on-demand billed in arrears honest structure
OpenAI Codex (ChatGPT Plus)$20/mo"expanded… Codex usage" · "Limits apply" (no number)
Devin / Windsurf Pro$20/mo"Increased quotas; quota refreshes daily and weekly" · up to 10 concurrent sessions
Gemini AI Pro (Antigravity)$19.99/mo"4x higher usage limits than Free" · higher Antigravity rate limits
Claude Max$100–$200/mo"5x or 20x Pro usage" (two tiers, no absolute number)
OpenAI Codex (ChatGPT Pro)$100–$200/mo"Maximum Codex tasks" · 5x or 20x Plus usage

Prices are canonical USD from each vendor's official US pricing page. Plans are grouped roughly by price, not ranked. "Sign-ups paused" reflects GitHub's own notice on the Copilot plans page on the date checked.

The honest winner: GitHub Copilot publishes real numbers

Copilot is the only plan here that gives you a figure you can actually budget against. The Free tier is 2,000 completions per month. The paid tiers are denominated in monthly credit dollars — $15 on Pro, $70 on Pro+, $200 on Max — so you can reason about how far a plan stretches before it meters out. That transparency is rare and genuinely useful.

The catch is large: GitHub states that new paid sign-ups are temporarily paused on Copilot Pro, Pro+, and Max. Existing Student/Pro/Pro+ customers can upgrade between tiers, but a new customer cannot buy a paid Copilot plan right now. So the plan with the best-published limits is, for most people reading this, not purchasable. The Free tier (2,000 completions) is still open.

The honest runner-up: Cursor tells you the shape of the limit

Among plans you can buy today, Cursor is the most straightforward about how the limit works, even though it doesn't print a number. Cursor Pro ($20/month) gives you an included model-usage pool, and once you exhaust it, on-demand usage is billed in arrears at cost. That is a different and more honest model than a hard cliff: you don't get cut off, you get charged. Pro+ ($60) and Ultra ($200) simply enlarge the included pool — Cursor describes Pro+ as "recommended for daily agent users" and Ultra as "for agent power users." You still can't compare its pool to Copilot's credits directly, but at least you know the mechanism.

Everyone else hides behind a multiplier

The rest describe coding limits in relative terms only. Claude bundles Claude Code into Pro ($17) and offers 5x or 20x on Max, but never states the base, so "20x" is 20 times an unknown. OpenAI Codex rides on ChatGPT Plus ($20, "expanded Codex usage", "limits apply") up to Pro ($100/$200, "maximum Codex tasks") with the identical 5x/20x ladder. Gemini grants "higher Antigravity rate limits" without saying the rate. Devin (formerly Windsurf) is slightly better — it tells you quotas "refresh daily and weekly" and that you get up to 10 concurrent sessions — which is shape, not size. Mistral's Vibe Pro ($14.99) promises "all-day coding" but caps it with "fair usage limits." None of these lets you answer "how many agent runs do I get?" before you pay.

So which should you actually pick?

If you only need autocomplete and occasional help, GitHub Copilot Free (2,000 completions/month) is the most generous free tier with a stated number. If you want a buyable plan with predictable economics, Cursor Pro ($20) won't cut you off mid-task — it meters overage instead of capping. If you already live in Claude or ChatGPT, Claude Pro ($17, includes Claude Code) and ChatGPT Plus ($20, Codex) fold coding into a plan you may already pay for, at the cost of never knowing the ceiling until you hit it. And if you're a heavy agent user, the whole category converges on the same $100–$200 5x/20x ladder — Claude Max, ChatGPT Pro, Copilot Max — where you're buying a bigger unknown for a known price.

The real lesson: in coding plans, the limit is the spec, and most vendors won't print it. The numbers that exist (Copilot's credits, Devin's refresh cadence, Gemini's Flow credits) change quietly between blog refreshes. QuotaLedger re-checks every one of these pages daily and timestamps the wording, so this comparison reflects what each vendor says on the day you read it. See the live ledger →

FAQ

Which AI coding plan has the best limits?
By transparency, GitHub Copilot — it publishes real numbers (2,000 completions on Free; $15/$70/$200 monthly credits on paid). But its paid sign-ups are paused, so among buyable plans, Cursor Pro is the most honest (included pool, on-demand billed in arrears rather than a hard cap).
Is Claude Code included in Claude Pro?
Yes — Anthropic states Claude Pro ($17/month annual, $20 monthly) includes Claude Code and Claude Cowork, subject to usage limits. Max (from $100) is 5x or 20x Pro usage.
What's the cheapest coding plan?
GitHub Copilot Pro is $10/month ($15 credits) but new paid sign-ups are paused; Mistral Vibe Pro is $14.99/month for "all-day coding"; Claude Pro is $17/month and includes Claude Code.
Why can't I sign up for GitHub Copilot Pro?
GitHub's plans page states new paid sign-ups are temporarily paused on Pro, Pro+, and Max. Existing Student/Pro/Pro+ customers can still upgrade; the Free tier stays open.
Does this change?
Constantly — coding limits and credit amounts are among the most-adjusted figures in AI pricing. QuotaLedger re-checks each vendor daily and logs changes with the date. See the live ledger and the changelog.
These figures are kept current automatically. QuotaLedger snapshots every vendor's pricing page daily, preserves the wording verbatim, and logs any change with a timestamp — so this explainer does not quietly go stale the way most do. See the full ledger →