Every tier, side by side
| Price | Claude | ChatGPT / Codex | Cursor | GitHub Copilot |
| Free | Free tier (chat, limited) | Free (limited Codex) | Hobby: limited agent, Tab | Free: limited |
| $8–10 | — | Go $8 | — | Pro $10 (~$15 credits) |
| $20 | Pro: 10–40 CC prompts/5 h, 40–80 Sonnet hrs/wk | Plus: 15–80 GPT-5.5 msgs/5 h | Pro: ~$20 included usage, unlimited Tab | — |
| $39–60 | — | — | Pro+ $60: 3× Pro | Pro+ $39 (~$70 credits, Opus-class) |
| $100 | Max 5x: 50–200 prompts/5 h, Opus | Pro 5x: 75–400 msgs/5 h | — | Max $100 (~$200 credits) |
| $200 | Max 20x: 200–800 prompts/5 h | Pro 20x: 300–1,600 msgs/5 h | Ultra: 20× Pro | — |
| Teams | $25 std / $125 premium per seat | Business $25 ($20 annual) | $40 per user | Business $19 / Enterprise $39 per seat |
| Overage | Usage credits at API rates | Codex credits (~4¢) | On-demand at API rates | 1¢ per credit |
Four different ways to sell the same thing
The tools look interchangeable at $20 and aren't. Claude sells a time-boxed allowance: a 5-hour window and a weekly cap, with the Max tiers as multipliers. You never see a dollar figure until you enable usage credits. Cursor sells an editor with a dollar amount of model usage baked in (about $20 on Pro), then bills on demand — transparent, but heavy agent users blow through the included amount in days. Copilot moved to a credit wallet in mid-2026: Pro's $10 buys about $15 of credits with completions free and unmetered, which makes it the best deal by far if autocomplete is most of what you want and a poor one if you run agents all day. Codex mirrors Claude's structure — Plus, Pro 5x, Pro 20x — with message counts per 5 hours rather than hours per week.
Who each one is cheapest for
- Autocomplete-first developers: Copilot Pro at $10. Unmetered completions, and the $15 in credits covers occasional chat. Nothing else comes close.
- One to three hours a day of agentic work: Claude Pro or Cursor Pro at $20, with Claude giving more model time and Cursor giving the nicer editor. Cursor's included usage runs out faster if you lean on the agent.
- Half a day, every day: the $100 tier. Claude Max 5x and ChatGPT Pro 5x include the frontier model; Copilot Max gives ~$200 of credits; Cursor Pro+ at $60 is the cheapest, but with 3× Pro's included usage it's the first to hit overages.
- All day with parallel agents: $200 — Claude Max 20x, ChatGPT Pro 20x or Cursor Ultra. At this level the deciding factor is which agent you trust, not price.
- Teams: Copilot Business ($19) for completions across an org; Claude Team premium ($125, ~Max 5x usage) or Cursor Teams ($40 + usage) for agent-heavy teams.
Combining tools
The most common real setup is two subscriptions: Copilot Pro ($10) for inline completions in the editor, plus Claude Max or Cursor for agentic work. That's $30 for a light user or $110 for a heavy one. It's a good deal until you start paying overages on both; if your combined bill passes $150, one $200 plan on the tool you use most is simpler and usually gives more usage.
Size my Claude plan precisely →
Frequently asked questions
Which AI coding tool is cheapest?
Copilot Pro ($10) for light/completion-first use; Claude Pro, Cursor Pro or ChatGPT Plus ($20) for moderate agentic use; $100 tiers for heavy daily use.
Cursor or Claude Code?
Cursor is the editor plus ~$20 of usage; Claude Pro is a weekly allowance of Claude Code time. Heavy agent users get more from Claude Max; light users often prefer Cursor.
What does Copilot cost now?
Pro $10 (~$15 credits), Pro+ $39 (~$70), Max $100 (~$200), Business $19/seat, Enterprise $39/seat; completions unmetered.
What's a Codex message limit?
Plus: 15–80 GPT-5.5 messages per 5 hours; Pro 5x: 75–400; Pro 20x: 300–1,600. More on GPT-5.4.
Are annual plans cheaper?
Claude Pro $17 vs $20; ChatGPT Business $20 vs $25; Cursor and Copilot offer annual discounts on some tiers.
Do free tiers work for real coding?
For trying things out. All of them throttle hard; Cursor Hobby and Copilot Free are the most usable for completions.