Prices and limits checked September 16, 2026

What does AI coding actually cost? Five calculators, current prices.

Plan limits and token prices change every few months, and every blog post about them is out of date within weeks. These tools carry the current numbers from the vendors' own pages, show the date they were checked, and do the arithmetic for your hours, your tokens and your budget.

Advertisement

The numbers, as of this week

ThingCurrent figure
Claude Pro$20/mo ($17 annual) · ~10–40 Claude Code prompts / 5 h · ~40–80 Sonnet hrs/wk · no Opus in Claude Code
Claude Max 5x / 20x$100 / $200 · ~50–200 / 200–800 prompts per 5 h · Opus included with its own weekly cap
Claude API (per M tokens)Sonnet 5 $2/$10 · Opus 5 $5/$25 · Fable 5.1 $10/$50 · Haiku 4.5 $1/$5 · cache reads 10% · batch 50% off
Claude Code on the API~$13 per developer per active day; $150–250/mo (Anthropic benchmark)
ChatGPT / CodexPlus $20 · Pro 5x $100 · Pro 20x $200 · GPT-5.5 $5/$30 per M
CursorPro $20 · Pro+ $60 · Ultra $200 · Teams $40/user
GitHub CopilotPro $10 · Pro+ $39 · Max $100 · Business $19 · Enterprise $39 (AI credits, completions unmetered)
Always-on agent$5–15 VPS + $1–6 tokens (light) to $100–200+ (heavy)

Why a calculator instead of an article

Claude's usage limits moved from a simple 5-hour window to 5-hour plus weekly caps; Copilot replaced premium requests with a credit wallet in June 2026; Cursor has repriced twice; OpenAI added Go and split Pro into 5x and 20x. Anything written down about "what it costs" is a snapshot. A calculator with a visible check date is a snapshot too — but one we can update in five minutes when a vendor moves, and one that does your specific arithmetic instead of a generic example.

Three things that matter more than the plan

Context size. Every AI coding tool resends your whole conversation on every request. Clearing between tasks is the difference between the top and bottom of every usage range. Model choice. Sonnet-class models handle most coding; frontier models cost 2.5–5× more per token and draw weekly limits proportionally faster. Caching. Stable prompts are served at a tenth of the price; anything that changes the system prompt on every call throws that away.

Independence

This site is not affiliated with Anthropic, OpenAI, Cursor, GitHub or any hosting company. Prices come from their public pages and are re-checked regularly; if you spot one that's moved, tell us and it'll be fixed the same day.