AI Coding Subscriptions Compared: Prices and Limits 2026

AI coding subscriptions compared on 4 October 2026: entry plans cost $10 to $20 a month, high-usage tiers cost $100 to $200, and open-model plans start at $1 to $10. GitHub, OpenCode, Command Code and Charm Hyper publish allowances you can convert to dollars. Anthropic, OpenAI, Google and Cursor describe theirs in multiples, message ranges or not at all, so we quote them as written.

How do AI coding subscriptions compare in one table?

Every figure comes from the vendor's own page, read on 4 October 2026. Prices exclude tax unless the vendor says otherwise. "Not published" means the vendor gives no number.

Plan	Price per month	Models	Usage limits, vendor wording	Works in	Terms note
Claude Pro	$20 ($17 billed annually)	Opus, Sonnet, Haiku; Fable on usage credits	"More usage" than Free, number not published	Claude Code, Claude apps	Sign-in is for Anthropic's own apps
Claude Max 5x, 20x	$100, $200	All Claude models; Fable up to 50% of weekly limit	5x or 20x Pro's per-session usage, plus a weekly limit	Claude Code, Claude apps	Same as Pro
ChatGPT Plus	$20	GPT-6.1 Sol, GPT-6 Luna	Local messages per 5 hours: Sol 15–160, Luna 350–3,000	Codex on web, CLI, IDE, iOS	Partner apps via Sign in with ChatGPT
ChatGPT Pro	$100, $200 or $500	Adds GPT-6 Astra	No five-hour limit now; tier multiples not published	Codex	Credits extend usage
Google AI Pro	$19.99	Gemini 3.1 Pro, 3.8 Flash in Antigravity	"Entry" limits, refreshed every five hours up to a weekly cap	Antigravity IDE and CLI, Jules	Gemini CLI ended for it on 18 June
Google AI Ultra	$99.99 (5x), $199.99 (20x)	Adds third-party models	5x or 20x Pro limits	Antigravity, Jules	No bring-your-own-key
GitHub Copilot Pro, Pro+, Max	$10, $39, $100	Claude, GPT, Gemini, Grok, Kimi K3 (Pro: a selection)	1,500, 7,000, 20,000 AI credits; 1 credit = $0.01	IDEs, Copilot CLI, GitHub, OpenCode	Unused credits lapse monthly
Cursor Pro, Pro Plus, Ultra	$20, $60, $200	Cursor models plus third-party at API rates	Two usage pools, amounts not published	Cursor editor, CLI, JetBrains	Extra usage at API rates
Devin Desktop (was Windsurf)	Free; $20 Pro; $200 Max	OpenAI, Claude, Gemini, SpaceXAI, open models	Daily and weekly allowance, not published	Devin Desktop, CLI	Extra usage at API pricing
OpenCode Go, Go Plus	$10, $40	About 30 models, mostly open	Monthly dollar limit per model; 20% per 5 hours, 50% per week	OpenCode, compatible agents	Traffic monitored for abuse
OpenCode Zen	Pay as you go	Claude, GPT, Gemini, open models	Your balance	Any agent	"Zero markups"; card fee passed on
Command Code Go, GOAT, Pro	$1, $10, $20 plus processing fee	Open models; Pro adds premium	$10, $70, $80 of credits, per-model allowances	Command Code CLI; API except Go	Sold only on its own site
Charm Hyper	Free; $20	23 coding models	100 Hypercredits a month free; 250 refreshing daily on $20	Crush, Claude Code, Codex, OpenCode, Pi, Zed	No training on your data

What does each plan give you, one by one?

Claude Pro and Max

Claude Code is included from Pro upward; the Free plan does not have it. Max includes Fable 5.1 as part of the plan, up to 50% of your weekly limit, while Pro pays for Fable with usage credits at API rates. Our subscription vs API guide works out the break-even.

ChatGPT Plus and Pro

Codex and ChatGPT Work share one allowance. OpenAI calls its message ranges estimates, not fixed limits, and lets you buy credits or switch to an API key at API prices. Pro currently has no five-hour limit, and OpenAI does not publish how the $100, $200 and $500 tiers differ in multiples.

Google AI Pro and Ultra

Coding moved to Antigravity. On 18 June 2026 Gemini CLI and the Code Assist extensions stopped serving free, AI Pro and AI Ultra accounts, and Antigravity CLI replaced them; Standard and Enterprise licences keep Gemini CLI. Google's plan page lists Ultra at $99.99 for 5x and $199.99 for 20x Pro's limits. Its Romanian page shows Pro at RON 109.99.

GitHub Copilot

Copilot now bills in AI credits. Pro's $10 buys 1,000 base credits plus a 500 flex allotment that GitHub calls variable. That is 1,500 credits, or $15 of usage at model prices (1.5x). Pro+ gives $70 for $39 (1.8x) and Max $200 for $100 (2x). Business and Enterprise carry no flex: 1,900 and 3,900 credits for $19 and $39. Code completions stay unlimited, and GitHub's changelog confirms that Pro, Pro+, Business and Enterprise work in OpenCode.

Cursor

Cursor splits usage into a pool for its own models and a pool for third-party models billed at API rates, and publishes no amounts. Its docs say daily Agent users typically use $60–$100 of total usage a month, so the $20 plan may need on-demand usage on top. Start, at ₹649, is for India only.

Devin Desktop, formerly Windsurf

Windsurf was renamed Devin Desktop. Its pricing page lists Free, Pro at $20 and Max at $200 with daily and weekly quotas it does not quantify. SWE-2 is free in Devin Desktop and the CLI through 16 October 2026.

OpenCode Go, Command Code and Charm Hyper

These publish the most numbers. OpenCode Go gives each model a monthly dollar limit: $10 buys up to $60 of list-price usage on cheap models and $15 on Kimi K3. Command Code GOAT's $10 buys up to $70 of credits on some models and $10 on Claude Sonnet 5.5. A Hypercredit is 5¢. Details and caveats are in our guide to the newer plans.

What does each plan return per dollar?

Where a vendor publishes dollar allowances, you can divide them by the price. This is our arithmetic on each vendor's own list prices, and it holds only for the models named.

• Copilot: Pro 1.5x ($15 of usage for $10), Pro+ 1.8x, Max 2x, Business and Enterprise 1x.
• OpenCode Go: 1.5x to 6x depending on the model ($15 to $60 for $10). Go Plus spans the same range ($60 to $240 for $40).
• Command Code: GOAT returns 1x to 7x ($10 to $70 for $10) and Go 6x to 10x ($6 to $10 for $1), before the processing fee.
• Charm Hyper: a prepaid bundle returns 1x ($20 buys 400 credits at 5¢ each). Whether the $20 subscription beats that depends on what "refreshing daily" means, which Hyper's pages do not spell out.
• Claude, ChatGPT, Google, Cursor, Devin: not computable from published numbers.

A multiple says how much list-price usage you can draw, not how good the answers are. A 6x plan on a cheap model and a 1.5x plan on a frontier model are different products.

Can you use a subscription in third-party tools?

Read the provider's terms every time, because they differ. Anthropic says OAuth sign-in is intended for its own plans and native apps, and it does not permit routing requests through Free, Pro or Max credentials on behalf of users; use an API key elsewhere. OpenAI lets partner apps use your ChatGPT limits with a cap you set. Z.ai restricts its plan to supported tools and bans account sharing. Never share or resell a subscription.

A gateway such as My Claude Code routes API-key traffic to many providers. Its README labels Claude subscription OAuth as not permitted by Anthropic, so keep subscriptions in the vendor's own clients.

What about GLM, Kimi, MiniMax and Qwen plans?

One line each; a separate article goes deeper.

• GLM Coding Plan: from $18 a month in three tiers. Z.ai's docs give 5-hour caps of 2,000, 12,000 and 28,000 credits, and weekly caps of 10,000, 60,000 and 140,000.
• Kimi Code: bundled with Kimi membership. Kimi's English help page lists ¥49 to ¥699 a month, and we could not verify a USD price.
• MiniMax Token Plan: Plus $22, Max $55 and Ultra $132, with 5-hour rolling and weekly windows.
• Qwen: the CLI is open source, but the free OAuth tier ended on 15 April 2026. Alibaba's Coding Plan Pro is $50 for up to 6,000 requests per 5 hours and 90,000 per month, with limited slots.

Which free tiers still exist?

Copilot Free has 2,000 completions and limited agent use with automatic model selection. Cursor Hobby and Devin Free have light quotas. Antigravity's free quota refreshes weekly, and Charm Hyper gives 100 Hypercredits (about $5) a month. The Gemini API free tier covers Gemini 3.8 Flash, but Google says free-tier data may improve its products. Two free routes are gone: Gemini CLI for individuals and Qwen OAuth.

Which plan should you pick?

Start from your tool, then your terms, then your volume. If you live in VS Code or on GitHub, Copilot fits. If you want a vendor's own agent, take that vendor's plan. If your volume is high and models are interchangeable, dollar-based open-model plans give the most predictable numbers. To price your own monthly calls, use our usage-based plan guide. Plans change often: Fable moved to usage credits in July, and Copilot swapped premium requests for AI credits around 1 June.

How does Danube Labs use this?

Danube Labs builds and runs LLM tooling daily, and My Claude Code exists because no single plan covers every tier of work. If you want help choosing models, budgets and guardrails for a team, see our AI systems service.