AI coding plans compared Claude Code, Codex, Cursor and Copilot
Paid coding plans start at $8 a month (OpenAI Codex Go) and go up to $500 for the heaviest individual tiers. Plan prices, published limits and what the same work costs on the API, checked 13 Oct 2026.
Plans for one developer
| Tool | Plans and prices per month | Beyond the allowance |
|---|---|---|
| Claude CodeAnthropic | Pro $20, Max 5x $100, Max 20x $200 | Usage credits or an API key, at API rates |
| OpenAI CodexOpenAI | Free; Go $8, Plus $20, Pro 100 $100, Pro 200 $200, Pro 500 $500 | Usage credits or an API key, at API rates |
| CursorAnysphere | Free; Pro $20, Pro+ $60, Ultra $200 | On-demand usage at API rates |
| GitHub CopilotGitHub | Free; Pro $10, Pro+ $39, Max $100 | Extra AI credits at $0.01 each |
US dollar list prices before tax, checked 13 Oct 2026. Each tool’s page has team prices, the published limits and sources.
What a coding agent costs on the API
| Model | Light5 tasks a day | Daily20 tasks a day | Heavy60 tasks a day | Per task |
|---|---|---|---|---|
| Claude Opus 5.5$4.00 / $20.00 | $213 | $853 | $2,560 | $1.94 |
| Claude Sonnet 5.5$2.00 / $10.00 | $107 | $427 | $1,280 | $0.97 |
| GPT-6.1 Sol$2.00 / $10.00 | $107 | $427 | $1,280 | $0.97 |
| Gemini 3.8 Flash$0.75 / $3.75 | $48.72 | $195 | $585 | $0.44 |
| GLM-5.3$1.40 / $4.40 | $115 | $458 | $1,375 | $1.04 |
| Claude Haiku 5.5$0.10 / $0.50 | $8.65 | $34.59 | $104 | $0.079 |
| GPT-6 Luna$0.10 / $0.50 | $6.50 | $25.98 | $77.95 | $0.059 |
Monthly cost of a coding agent doing tasks of 40 steps each (an 18K-token system prompt, 90% of repeated input cached), on 22 working days a month. A plan is worth it when its limits cover more work than its price buys here; providers don’t publish limits in tokens, so check your own usage. Change the workload
Other coding tools
| Tool | Price | What’s included | Beyond the allowance |
|---|---|---|---|
| Devin Desktop (formerly Windsurf) | Free; Pro $20; Max $200; Teams $80 plus $40 per seat | Daily and weekly quotas (not quantified) | Extra usage at API prices Source |
| JetBrains AI | Pro $10; Ultimate $30; Enterprise $60 per user | 10 or 35 credits per 30 days; includes Junie, Claude Agent and Codex | Top-ups at $1 per credit Source |
| Google Antigravity and Jules | Free; higher limits with Google AI Pro ($19.99) or Ultra | Jules: 15, 100 or 300 tasks a day on Free, Pro and Ultra | Pro and Ultra get a flexible AI credit pool Source |
Checked 13 Oct 2026.
How to choose
- You want the strongest models in a terminal agent: Claude Code on Claude Max, or Codex on ChatGPT Pro. Both raise the limits a long way at $100–200 a month.
- You want one editor with every provider’s models: Cursor or GitHub Copilot. Copilot is cheaper to start, and its credits are metered in dollars, so you can see what you use.
- You already pay for ChatGPT or Claude: Codex and Claude Code are included in Plus and Pro, so try them before buying another tool.
- You run agents in CI or scripts: use an API key and per-token billing. Plans are for people, not automation.
Questions
Which AI coding plan is cheapest?
OpenAI Codex Go, at $8 a month, is the cheapest paid plan of the four. What it includes: GPT-6 Luna at standard speed in the desktop app (rolling out). OpenAI Codex, Cursor and GitHub Copilot have free tiers with tight limits.
Is a coding plan cheaper than the API?
For one developer using an agent every day, often. 20 agent tasks a day costs about $427 a month on Claude Sonnet 5.5 at API rates, far more than a $20 plan, but that plan’s limits won’t stretch that far. For occasional use or automation, the API can be cheaper.
Do the plans publish their limits in tokens?
No. Anthropic, OpenAI and Cursor publish limits as multiples of another plan or as message estimates. GitHub Copilot is the exception: its AI credits are metered by token at $0.01 a credit.
When were these prices checked?
On 13 Oct 2026, on each tool’s own pricing page. Every plan links to its source.