AI Coding Agent Pricing Explained: Claude Code, Cursor, Codex and Copilot (2026)

Coding agent prices look simple (most start at about $20 a month) and are not. What you actually pay depends on how much model usage the plan includes, what happens when you run out, and whether the tool bills you per token instead. This guide explains the pricing models and lists what each major agent charges. Prices are read from the vendors' pricing pages; each agent's page shows when they were last checked.

The big four

• Claude Code has no free tier. It comes with Claude's paid plans, and heavier users move to the Max tiers for higher limits. It can also run on API billing, paid per token.
• Cursor has a free Hobby plan and paid tiers that differ mainly in how much model usage is included; teams pay per seat.
• OpenAI Codex is included in ChatGPT plans, from the free plan upward, with limits that grow with the plan.
• GitHub Copilot has the cheapest paid entry point and meters agent usage in credits shared across chat, the CLI and the cloud agent.

Side by side on usage and features: Claude Code vs Codex vs Cursor.

The three ways you can be billed

1. Subscription with usage limits. A flat monthly price includes a quota of model usage, reset every few hours or each week. This is the best deal for steady daily use: heavy users get far more model time than the same money would buy through an API. The catch is the cap. When you hit it, you wait, upgrade to a higher tier or switch to slower models.

2. Credits. The plan includes a number of credits and each request or each minute of agent work uses some. Kiro, Factory's Droid and Copilot work this way. Credits make costs visible, but a credit buys a different amount of work in each product, so compare plans within one vendor, not across vendors.

3. Bring your own key (pay per token). Open-source agents such as OpenCode, Cline and Aider are free; you pay the model provider directly. A light week costs a few dollars. A week of long autonomous sessions on a top model can cost more than a premium subscription, because agents send a lot of context back to the model on every step.

Other agents and what they charge

Estimating your monthly cost

• Occasional use (a few tasks a week): a free tier, a $10 to $20 plan, or an open-source agent with your own key.
• Daily use by one developer: the $20 tier of the tool you like most. Watch whether you hit the limit; if you do most weeks, the next tier is usually cheaper than paying for overage.
• All-day agent use or several agents in parallel: the $100 to $200 tiers that most vendors now offer. At this level the flat price is usually cheaper than API billing.
• A team: per-seat business plans add admin controls, data-retention terms and pooled usage. Check whether usage is pooled or per seat: pooled usage lets light users subsidise heavy ones.

Three questions before you pay

1. Which models are included, and which cost extra? The headline model is often limited on the entry tier.
2. What happens at the limit? Hard stop, slower model, or automatic overage billing? The last one is the only way to get a surprise bill.
3. Can you use your own key? A tool that accepts your API key gives you a way out if the plan's limits do not fit.

For free options, see free AI coding agents; for the full comparison by usage, best AI coding agents. Prices on this page update automatically when the vendors' pages change; the pricing overview covers every agent we track.