What coding agents cost: Claude Code, Codex, Cursor and API usage

Prices last checked on 6 November 2026 from each vendor's own pricing page, linked under Sources. Prices are in US dollars before tax. Vendors change these often, so check the source before you buy.

Subscription plans

Free tiers exist too. Claude Free does not include Claude Code. ChatGPT Free includes limited Codex use. Cursor Free includes limited agent requests.

For teams: Claude Team seats cost $20 a month billed yearly or $25 monthly, and premium seats with 5x the usage cost $100 or $125. ChatGPT Business costs $20 per user a month billed yearly, or $25 monthly. Cursor Teams starts at $30 per user a month.

What usage limits mean

None of these plans sells a fixed number of tasks. Each one gives you an allowance that a long task with a big context uses up much faster than a quick fix.

• Claude plans reset every 5 hours and every week. How fast you reach a limit depends on the model and the work.
• Codex publishes estimates per 5 hours for each model on Plus. GPT-6.1 Sol is 15 to 160 local messages, and GPT-6 Luna is 350 to 3,000. OpenAI says these are estimates and not fixed limits, and that weekly limits may also apply.
• Cursor gives each plan a pool shared between Cursor and Grok that refreshes weekly. Other vendors' models are billed on top at API list price.

Running several agents at once uses your allowance faster. Claude Code's docs say it plainly: running several sessions or subagents at once multiplies token usage.

API prices

With an API key you pay for every token, with no monthly cap. These are list prices at standard speed.

OpenAI's batch prices are half of these, and its Fast and Ultrafast modes cost more. Cursor charges other vendors' models at their API list price. Cached input is the price of reading a cached prompt; writing to the cache costs extra.

Subscription or API key?

For one person coding most days, a subscription usually costs less. Anthropic's own figures put the average enterprise Claude Code user at $150 to $250 a month at API rates, which is more than a Max plan, with 90% of users under $30 per active day.

API keys make sense for CI jobs, shared automation and teams that need per-project billing. OpenAI describes the Codex API key option as a fit for automation in shared environments like CI, and it comes without Codex's cloud features.

Use the plans you already pay for

Most people we talk to already pay for a Claude or ChatGPT plan, often both. Before adding a third subscription or an API key, get the most out of those.

• Sign each CLI in with your subscription. claude asks you to log in on first use, and codex offers Sign in with ChatGPT. If ANTHROPIC_API_KEY is set and you approve it, Claude Code uses the key and bills per token instead, so check which one is active before a long run.
• Each plan has its own allowance, so spreading work across vendors helps. A Claude Code task and a Codex task running together draw on two separate limits.
• Use smaller models for small jobs. On ChatGPT Plus, OpenAI's estimates run from 15 to 160 messages per 5 hours with GPT-6.1 Sol and from 350 to 3,000 with GPT-6 Luna.
• Claude Code warns you when resuming a long conversation after a long break, because the saved conversation counts against your 5-hour limit.

Zoltbot is built around this. It runs Claude Code, Codex, Cursor and Antigravity through their own CLIs with the logins you already have, never copies your credentials, and charges nothing for tokens. It never resends chat history either: each tool keeps its own caching and compaction, and Zoltbot shows every token it adds. Zoltbot is free for individuals.