Why these pricing numbers matter

In 2026 the headline subscription price on an AI coding tool tells you almost nothing about what you will actually pay. Cursor, Claude Code, GitHub Copilot, and OpenAI's Codex have all converged on the same model: a monthly fee that buys you a pool of credits, and consumption metered against the underlying API token rates. The plan name is a deposit; the token meter is the bill. I build token tooling for a living, and the question I get most often is "which tool is cheapest?" — which is the wrong question. The honest answer is that on a given model, the per-token rate is identical no matter which wrapper you run it through, because they all bill against the same provider price sheet. What varies is how much of that you get included before the meter starts, and how efficiently the tool spends your tokens. This page collects the real June 2026 sticker prices and the token rates they meter against, so you can reason about cost instead of guessing. If you want the raw per-model API rates in one place, see LLM API Token Pricing. For how those rates translate into the cost of an actual coding session, see AI Coding Agent Token Costs.

Key Takeaways

•Cursor runs $20 Pro / $60 Pro+ / $200 Ultra per month, each with a metered credit pool against model API rates [1]
•Claude Code is $20 Pro / $100 Max 5x / $200 Max 20x, each a token budget that throttles when exhausted [2]
•GitHub Copilot moved to usage-based AI Credits on June 1, 2026 — Pro $10, Pro+ $39, billed on input, output, and cached tokens [3]
•Codex ships inside ChatGPT Plus ($20) and Pro ($200) and switched to API-style token billing on April 2, 2026 [4]
•The same model costs the same per token everywhere — Opus 4.8 is $5/$25 per MTok whether you reach it through Cursor, Claude Code, or the raw API [5]

1. What does Cursor cost per month in 2026?

Cursor's paid tiers are $20 (Pro), $60 (Pro+), and $200 (Ultra) per month, with a free Hobby plan and annual billing taking 20% off. [1] Each paid tier maps to a usage multiplier: Pro includes a roughly $20 credit pool against premium models, Pro+ gives 3× usage, and Ultra gives 20× usage across OpenAI, Anthropic, and Google models. [1] The important detail is buried in "credit pool." The Pro fee is not a flat all-you-can-eat licence — it is a prepaid balance metered against the same model rates the providers publish. Once you exhaust it, you are paying marginal token cost. So the practical cost of Cursor Pro for a heavy user is the $20 plus whatever overage their agent loop burns, and the agent loop's appetite is a function of how much context it ships on every turn.

2. What does Claude Code cost per month?

Claude Code is $20 on the Pro plan, $100 on Max 5x, and $200 on Max 20x, or pay-per-token directly through the Anthropic API with no monthly minimum. [2] The Pro plan grants access to both Sonnet 4.6 and Opus, with a token budget sized for "focused coding sessions"; Max 5x and Max 20x raise that budget 5× and 20× respectively. [2] The word doing the work in that sentence is "budget." None of these tiers is unlimited — each is a metered allowance, and when you hit the ceiling you get rate-limited until the window resets. That is why two developers on identical $100 Max plans can have wildly different experiences: the one whose agent re-reads a 50k-token file on every turn exhausts the budget far sooner than the one whose tool only sends the relevant function. The budget is fixed; the burn rate is not.

3. What does GitHub Copilot cost now that billing changed?

GitHub Copilot is $10/month (Pro) and $39/month (Pro+) for individuals, with Business and Enterprise at $39/user/month — and as of June 1, 2026 it runs on usage-based billing. [3] Instead of counting "premium requests," every plan now includes a monthly allotment of GitHub AI Credits, and usage is calculated on token consumption — input, output, and cached tokens — at the listed API rate for each model. [3] This is the clearest statement of the trend on this page: the largest AI coding tool by install base just told its users, explicitly, that the bill is now a function of tokens. The plan fee buys credits; the credits drain at the model's published rate. Copilot's own billing copy now reads like an API invoice, which is the right mental model for all four of these tools.

4. Is Codex cheaper than the others?

Codex is included in ChatGPT Plus ($20/mo) and Pro ($200/mo), and on April 2, 2026 OpenAI moved Codex from per-message pricing to API-style token rates. [4] Credits are still what you buy, but consumption now depends on input, cached input, and output tokens, the same way the API meters. OpenAI's own guidance puts typical Codex usage at roughly $100–$200 per developer per month depending on model, instance count, and how much "fast mode" you run. [4] So the answer to "is Codex cheaper?" is: not structurally. At $20 the entry fee matches Cursor Pro and Claude Code Pro almost exactly, and once you are token-metered the cost is GPT-5.5's rate of $5.00 input / $30.00 output per MTok, which is, if anything, a higher output premium than Claude. [6]

5. Why the per-token rate is the real comparison

The same model costs the same per token regardless of which tool you reach it through, because every tool here meters against the provider's published API price sheet. [5][6] Those rates are the constant the whole market shares:

Model	Input $/MTok	Output $/MTok	Cache Read $/MTok
Claude Opus 4.8	$5.00	$25.00	$0.50
Claude Sonnet 4.6	$3.00	$15.00	$0.30
Claude Haiku 4.5	$1.00	$5.00	$0.10
GPT-5.5	$5.00	$30.00	$0.50

Cache reads are priced at ≈10% of base input — a 90% discount on repeated context. Sources: [5][6], June 2026.

This is why I keep saying the cheapest tool is a category error. If two tools both run Opus 4.8, the one that wins on cost is the one that sends fewer tokens to reach the same answer — the one that doesn't re-paste your whole repo into context, uses prompt caching for the parts that never change, and trims tool output before it lands in the window. The plan fee is a rounding error next to a chatty agent loop.

6. How much can you actually cut?

Because every plan on this page now meters against token volume, anything that reduces the tokens an agent sends and receives applies a flat percentage discount to your bill, on whatever tool and model you run. That is the lever, and it is provider-agnostic. Concretely: an agent that re-reads a 50k-token system context on every turn pays $0.25 per turn in input on Opus 4.8. Enable prompt caching and that drops to $0.025 per turn after the first write — a 90% cut on that slice. Stop the agent re-reading whole files it doesn't need, and you cut the slice itself. None of this requires switching tools or models; it requires sending less. This is the entire premise of what I work on. Tokenade sits at the proxy layer in front of Claude Code, Cursor, Codex, Copilot, and the rest, and trims the tokens going through: semantic code search so the agent reads only relevant code instead of whole files, output filtering on noisy tool results, skeleton compression for large files, lazy MCP loading, and a dashboard that shows the tokens and dollars saved. It is source-available under MIT — the free tier covers up to ≈10M tokens/month, Pro is $19.90/month excl. tax (US) / €19.90/month TTC (FR) with unlimited machines, and Enterprise (unlimited tokens, fleet-sized machine count) is custom-priced. For the practical playbook, see reducing Claude Code token usage and reducing AI coding agent token usage.

Source notes

Sources and references

[1]Cursor. "Pricing — Hobby, Pro, Pro+, Ultra". cursor.com/pricing, June 2026. Link ↗
[2]Anthropic. "Plans & Pricing — Claude (Pro, Max, API, Teams)". claude.com/pricing, June 2026. Link ↗
[3]GitHub. "GitHub Copilot · Plans & pricing" and "GitHub Copilot is moving to usage-based billing". github.com, June 2026. Link ↗
[4]OpenAI. "Pricing — Codex" and "Using Codex with your ChatGPT plan". developers.openai.com/codex/pricing, June 2026. Link ↗
[5]Anthropic. "API Pricing — Claude models". anthropic.com/pricing, June 2026. Link ↗
[6]OpenAI. "OpenAI API Pricing — GPT-5.x". openai.com/api/pricing, June 2026. Link ↗

Subscription tiers and the underlying API token rates are USD list prices as of June 2026; FR Tokenade pricing is quoted TTC. All four tools meter usage against the providers' published per-token rates — re-verify both the plan fee and the model rate before building a cost model, as both change at each release and billing-model update.

AI Coding Tool Pricing: Cursor vs Claude Code

Key figures