An AI coding subscription can cost $20 a month and run out of included usage before the next billing date. Claude Code and Codex have account-based limits, GitHub Copilot meters agent work through monthly AI Credits, and Cursor shares a weekly AI allowance with Grok. Coding sessions, model choice, and extra usage charges all affect what the monthly fee actually buys.
The subscription also determines where you can code. GitHub Copilot runs in IDEs and through Copilot CLI. Cursor and Devin Desktop are code editors with coding agents, and Devin also runs cloud agents. GLM, QwenCloud, and MiniMax subscriptions connect to third-party coding agents through their subscription keys or endpoints.
This comparison examines 16 AI coding subscription and token-plan families, including ChatGPT with Codex, Claude Code, GitHub Copilot, Cursor with xAI, Devin, and lower-priced multi-model options. It shows which plans offer coding agents, how allowances reset, which external clients work with each plan, and when charges increase.
Last updated: November 10, 2026.
AI coding plans by budget and use case
| Need | Start with | Key reason |
|---|---|---|
| $0 first | Codex, Antigravity, xAI Free | Limited coding access |
| Coding agent near $20 | ChatGPT Plus, Claude Pro | Codex or Claude Code |
| IDE coding | GitHub Copilot Pro | $10/month, monthly AI Credits |
| Editor and cloud agents | Devin Desktop | Free or $20/month Pro |
| Cursor editor | xAI Plus / Premium | One weekly AI pool |
| Browser and agent control | Google AI Pro / Ultra | Antigravity agents |
| Compatible coding agents | GLM, BytePlus | Subscription endpoints |
| Monthly credits | QwenCloud, Tencent | Monthly credit pools |
| Low-cost model | OpenCode Go | From $10/month |
| Coding and media | MiniMax M Plan | Coding agents and media credits |
| No weekly quota | QwenCloud, Xiaomi MiMo | Monthly credits |
AI coding plan prices and usage limits
Prices are in USD per month unless noted. Promotional and annual-billing rates are identified where used. Taxes and regional pricing may vary.
| Plan | Price / month | How usage is limited |
|---|---|---|
| ChatGPT + Codex | $0 / $8 / $20 / $100 / $200 / $500 | Work/Codex shared; Chat joins Dec (Plus/Pro/Business) |
| Claude + Claude Code | $20 Pro / $100 Max / $200 Ultra | 5h + weekly; verified Ultra: weekly only |
| Google AI Pro / Ultra | $19.99 / $100 / $200 | 5h and weekly caps, paid top-ups |
| xAI: Cursor + Grok Build | $0 / $8 / $30 / $100 / $200 | Weekly AI pool across Grok and Cursor |
| Mistral Vibe | $0 / $14.99 | Fair-use limits and paid overages |
| Kimi Code | ¥99 / ¥199 / ¥699 / ¥1,399 (Code included) | Shared monthly credits + rolling 5h cap; no weekly cap on new plans |
| GitHub Copilot | $0 / $10 / $39 / $100 | Monthly AI Credits: 1.5K / 7K / 20K on paid tiers |
| Devin Desktop (formerly Windsurf) | $0 / $20 / $200 | Paid agent usage resets daily and weekly |
| GLM Coding Plan | $18 / $80 / $168 | 5h + weekly credits (Lite 2K / 10K) |
| BytePlus ModelArk | $10 / $50 | 5h / weekly / monthly shared request quotas |
| QwenCloud Token Plan | $6 / $10 / $18 / $68 (promo) | 11.5K / 25.5K / 45K / 180K credits per subscription month |
| Tencent Token Plan | $7 / $17 / $51 / $103 | 1K / 2.6K / 7.9K / 15.9K credits/month |
| MiniMax M Plan | $22 / $55 / $132 | Shared model usage by tier; optional credit packs |
| Xiaomi MiMo Token Plan | $6 / $16 / $50 / $100 | 4.1B / 11B / 38B / 82B credits/month |
| Ollama | Pay as you go / $20 / $100 | Cloud pay-per-token; Pro $60, Max $300 credits/month |
| OpenCode Go / Go Plus | $10 / $40 | Per-model 5h, weekly, and monthly allowances |
Free AI coding access before you pay
| Free option | Access | Main limit |
|---|---|---|
| ChatGPT Free | Codex in the desktop app | GPT-6 Luna, limited use |
| GitHub Copilot Free | IDE agent, Copilot CLI, completions | Limited AI usage, 2,000 completions/month |
| Devin Free | Desktop editor, coding agents | Light agent quota, restricted model catalog |
| Antigravity Free | Editor, terminal, browser | Lower usage limits |
| xAI Free | Grok Build, Cursor, chat | Limited agent requests |
| Mistral Free | Vibe CLI and web | Limited coding sessions |
| MiniMax Code Free | MiniMax Code app and CLI | Limited M3.1 Flash; no external-agent Subscription Key |
| Ollama (local) | Models run on your hardware | No subscription; cloud use is billed per token |
| NVIDIA NIM | Selected hosted model endpoints | Availability and usage depend on endpoint |
| OpenRouter Free | Free model catalog | 50/day, 1,000/day after $10 purchased |
Chat subscriptions that include AI coding agents
These subscriptions bundle coding access with consumer AI accounts. Coding usage follows each provider’s subscription limits and billing rules.
ChatGPT with Codex
Official plans: Codex pricing and ChatGPT plan access
ChatGPT Free grants limited Codex access, Go costs $8 per month, and Plus costs $20. Pro costs $100, $200, or $500 depending on the tier. Free and Go use GPT-6 Luna at Standard speed in the desktop app, subject to rollout. Plus offers GPT-6.1 Sol and GPT-6 Luna in Codex. The $500 Pro tier adds Ultrafast mode.
As of November 2026, Codex and ChatGPT Work draw from the ChatGPT subscription’s agentic allowance. Regular Chat uses its own message limits. With Plus, you can use Codex through the CLI, IDE extension, web, and cloud code review. Pro tiers increase agentic capacity.
Starting in December 2026, Plus, Pro, and Business are scheduled to share a token-measured allowance across ChatGPT and Codex. Model choice and task complexity will determine how much of that allowance is used. Instant Chat will not draw from it. The combined remaining usage and reset times will appear in Settings → Usage. OpenAI has not published uniform token quotas or task-conversion figures for the planned shared system.
Sign in with ChatGPT to use the subscription allowance. OpenAI API requests incur token-based charges on the API billing account and do not unlock subscription-based cloud integrations. If your account offers credit purchases, you can buy more Codex usage after exhausting your plan’s allowance. Under the planned December system, your other options will be to wait for the reset or upgrade your plan.
Read More: OpenAI Codex Commands Cheat Sheet
Claude Pro, Max, and Ultra with Claude Code
Official plans: Claude Pro, Max, and Ultra pricing
Claude Pro, Max, and Ultra cost $20, $100, and $200 per month, respectively. Claude Code access starts with Pro. Pro is subject to five-hour and weekly limits. Max raises both allowances fivefold. Ultra raises the weekly allowance tenfold and removes the five-hour cap for verified accounts. Unverified accounts have a temporary session limit.
Claude Code offers Opus 5.5 for complex coding and Claude Fable 5.2 for especially demanding reasoning and long-horizon agentic work. Fable 5.2 requires Claude Code 2.1.321 or later. Sign in with your Claude account to use subscription allowances across Claude chat, Desktop, and Claude Code. Outside the launch promotion, Fable models run on usage credits for Pro and Team Standard; Max and Ultra include Fable models within up to 50% of their existing weekly allowance.
Fable 5.2 launch offer: From November 9 through November 16, 2026 at 11:59 p.m. PT, eligible Pro and Team Standard users can use Fable 5.2, 5.1, and 5 within up to 50% of their weekly allowance at no additional charge. Eligible existing subscribers also receive one Fable limit reset. Max, Ultra, Team Premium, and qualifying legacy Enterprise Premium users receive one to three resets based on paid subscription history, expiring March 8, 2027. The official offer conditions specify eligibility and reset deadlines.
Monthly Claude Platform API credits are $100 on Max and $200 on Ultra. After seven days on either plan, you can claim the credits by linking a Claude Console organization. The credits pay for Claude API calls, Console Playground, and Agent SDK usage. They cannot pay for interactive Claude Code sessions or subscription overages. When your Claude Code allowance runs out, you can continue at standard API rates.
Read More: The Ultimate Claude Code Resource List: Agents, Skills, Plugins & More
Google AI Pro with Antigravity
Official plans: Google AI plans
Google AI Pro costs $19.99 per month in the US. The $100 and $200 Google AI Ultra tiers offer approximately 5x and 20x the Pro Antigravity allowance. Antigravity runs coding agents in its desktop app, CLI, and browser, with controls for parallel tasks.
Antigravity quotas vary with task complexity, model usage, and account tier. On Pro and Ultra, Gemini quotas reset every five hours and have weekly caps. Other models use their own quotas. On a paid plan, you can purchase AI credits after reaching your Antigravity quota. Account and country restrictions apply.
Read More: Antigravity CLI Cheatsheet: Commands, Setup, Shortcuts, and Workflows
xAI subscription with Grok Build and Cursor
Official plans: xAI Free, Plus, Premium, Super, and Ultra
xAI uses one subscription for Grok, Grok Build, and Cursor. Free costs $0 with limited agent requests. Plus costs $8 per month, Premium $30, Super $100, and Ultra $200. Premium, Super, and Ultra multiply the Plus usage pool by four, 15, and 50, respectively. These multipliers do not represent a fixed number of coding tasks.
Grok Build runs in a terminal or headless CLI and handles project files, diffs, worktrees, subagents, skills, plugins, hooks, and MCP servers. Cursor is an editor with a coding agent. Premium and higher tiers unlock cloud agents, automations, Origin code hosting, and Bugbot code review. Plus grants Cursor access to Grok models. Premium and higher tiers also offer third-party models.
One weekly AI allowance pays for Grok Chat, Build, Imagine, Voice, Bot, Grok API access, and Cursor. Grok model requests in Cursor draw from that pool. Third-party models in Cursor incur charges at their API list prices. You can buy top-ups when your weekly allowance runs out.
Read More: Grok Build: SpaceXAI’s Open-source Terminal AI Coding Agent
Mistral Pro with Vibe
Official plans: Mistral Free and Pro pricing
Mistral Free offers Vibe coding sessions subject to usage limits. Pro costs $14.99 per month and raises the Vibe allowance for CLI, IDE, and web use under fair-use limits. Mistral Free has no advertised monthly API-credit allowance.
Pro comes with $25.50 per month in API credits at Mistral’s published USD price. If you enable pay-as-you-go credits, you can continue using Vibe at API rates after your subscription allowance runs out. Vibe also runs remote coding tasks alongside local CLI and IDE sessions.
Read More: Mistral Vibe: Open-Source CLI Coding Agent and Claude Code Alternative
Kimi membership with Kimi Code
Official plans: Kimi membership pricing
Kimi offers five membership tiers in China: Go at ¥49 per month, Plus at ¥99, Pro at ¥199, Max at ¥699, and Ultra at ¥1,399. Kimi Code and Agent Swarm start at Plus. Agent Credits scale from 2x the Go baseline on Plus to 28x on Ultra. Max and Ultra support up to four parallel Agent tasks. Andante, Moderato, Allegretto, and Allegro are legacy memberships for earlier subscribers.
Kimi Code draws monthly membership credits from the pool used by Kimi Work, Kimi Claw, Deep Research, slides, and other member features. The new Plus, Pro, Max, and Ultra tiers have a rolling five-hour Kimi Code limit with no weekly cap. Legacy plans retain both five-hour and weekly Kimi Code limits. K3.1 Pro starts at Plus with up to 256K context. Pro and higher tiers raise context to 1M and add K3.1 Pro Highspeed. When your Kimi Code allowance runs out, you can enable Extra Usage to pay for additional requests from a purchased balance.
Read More: Kimi Code CLI: Open-Source AI Coding Agent with Skills & MCP Support
IDE and cloud-agent coding subscriptions
GitHub Copilot Free, Pro, Pro+, and Max
Official plans: GitHub Copilot plans and pricing
GitHub Copilot Free offers limited chat and agent usage, Copilot CLI, and 2,000 code completions per month. Pro costs $10 per month, Pro+ costs $39, and Max costs $100. All three paid tiers provide unlimited inline completions and next-edit suggestions, plus Copilot cloud agents and code review. Copilot runs in supported IDEs, GitHub, and the terminal.
GitHub AI Credits pay for chat, agent mode, cloud agents, Copilot CLI, and other metered features. Monthly AI Credits are 1,500 ($15 usage value) on Pro, 7,000 ($70) on Pro+, and 20,000 ($200) on Max. Each AI Credit represents $0.01 in usage value. The totals comprise a base allocation and a flexible allowance that GitHub can adjust. Credits reset on the first day of the calendar month, regardless of the subscription renewal date. Model prices and token consumption determine usage. On a paid plan, you can buy more AI Credits when your monthly balance runs out.
GitHub replaced premium-request billing with AI Credits for individual usage-based plans in June 2026. Some existing annual subscriptions retain legacy premium-request billing until their current terms end.
Read More: 10 Best CLI AI Coding Agents in 2026: Open-Source & Commercial
Devin Desktop and Devin Cloud (formerly Windsurf)
Official plans: Devin individual pricing
Devin Desktop is the new name for the Windsurf editor. The Free tier limits agent usage and model access, with unlimited inline edits and Tab completions. Pro costs $20 per month and adds frontier models and Devin Cloud agents. Max costs $200 and raises agent usage limits. Devin Desktop and Devin CLI handle local coding tasks. Devin Cloud runs delegated coding jobs.
Paid agent usage has daily and weekly refresh windows. The number of tasks depends on model choice, session length, and task complexity. Devin does not publish a fixed request count for every tier. On Pro or Max, you can buy additional usage at API rates after reaching your agent allowance. Existing Windsurf subscriptions retain their previous plan pricing when moving to Devin Desktop.
Dedicated AI coding plans for external coding agents
Depending on the provider, a subscription key or API endpoint connects the plan to coding agents such as Claude Code, Codex, OpenCode, Cline, Cursor, or OpenClaw.
GLM Coding Plan
Official plan: GLM Coding Plan pricing and access
GLM Coding Plan costs $18 per month for Lite, $80 for Pro, and $168 for Max. Annual billing lowers the effective monthly prices to $12.60, $56, and $117.60. Lite provides 2,000 Credits per rolling five hours and 10,000 per week. The limits rise to 12,000 and 60,000 on Pro, and 28,000 and 140,000 on Max. Both time windows apply. Every tier offers GLM-5.3 and GLM-5.3-Flash.
The GLM coding endpoint works with more than 20 agent clients, including Claude Code, Codex, OpenCode, OpenClaw, and Cline. Input, cached input, and output tokens consume Credits, as do the Coding Plan’s MCP tool calls. Off-peak model calls consume half the standard Credits rate. GLM personal plans permit interactive agent use subject to the plan’s usage policy.
BytePlus ModelArk Coding Plan
Official plan: ModelArk Coding Plan
BytePlus ModelArk Coding Plan costs $10 per month for Lite and $50 for Pro. Lite has five-hour, weekly, and subscription-month quotas, with Pro providing five times the Lite allowance. BytePlus estimates approximately 1,200 requests every five hours, 9,000 per week, and 18,000 per subscription month for Lite when using dola-seed-2.0-lite in Claude Code. The corresponding Pro estimates are 6,000, 45,000, and 90,000. These are model-specific estimates, not guaranteed request counts. The coding model catalog includes GLM-5.3-Flash, DeepSeek V4.1 Flash, and Dola-Seed-2.0 Pro, with Auto selection available.
The endpoint works with Claude Code, Cursor, Cline, Kilo Code, Roo Code, and OpenCode. BytePlus stops plan usage when the quota is exhausted. Another package or account balance is not charged automatically.
QwenCloud Token Plan
Official plan: QwenCloud Token Plan Individual
QwenCloud Token Plan Individual now has four tiers. Promotional monthly prices are $6 for Lite, $10 for Essential, $18 for Standard, and $68 for Pro. Their monthly quotas are 11,500, 25,500, 45,000, and 180,000 Credits respectively. The old seven-day quota was replaced by a subscription-month allowance with no weekly quota limit.
Qwen, GLM, and DeepSeek requests consume Token Plan credits when sent through a coding client configured for QwenCloud. Purchase and use require QwenCloud’s Singapore region, an sk-sp- subscription key, and the token-plan.maas.qwencloudapi.com endpoint. Ordinary pay-as-you-go API keys cannot spend Token Plan credits. Credits reset on the subscription anniversary and expire if unused. Requests pause at the quota limit unless you upgrade or buy a Credit Pack. You can configure your personal key on multiple devices.
Tencent Cloud Token Plan
Official plan: Tencent Cloud Token Plan overview
Tencent Cloud Token Plan has four monthly Credits tiers: Lite at $7 for 1,000 Credits, Standard at $17 for 2,600, Pro at $51 for 7,900, and Max at $103 for 15,900. The model pool offers MiniMax, Kimi, GLM, and DeepSeek options.
The subscription works with Claude Code, OpenCode, Cline, Cursor, Kilo Code, Codex CLI, and more. Credits reset by subscription month. Personal plans permit only your own interactive coding sessions. Automation scripts, application backends, and shared accounts require another billing arrangement.
MiniMax M Plan
Official plan: MiniMax M Plan pricing and features
MiniMax M Plan costs $22 per month for Go, $55 for Explore, and $132 for Build. All three tiers offer MiniMax Code, the M3.1 Flash text model, image and speech generation, web search, multimodal understanding, and a Subscription Key for third-party coding agents. Explore has three times the Go allowance, and Build has 7.5 times. H3 video generation starts at Explore.
The M Plan Subscription Key connects Claude Code, Codex, Cursor, OpenCode, and OpenClaw. Requests count against the M Plan allowance. If you exhaust the M Plan allowance, you can buy credit packs for additional requests. The key never debits the normal pay-as-you-go API balance. MiniMax Code and M3.1 Flash offer limited free use. The free tier does not provide a Subscription Key for external coding agents. The older Plus, Max, and Ultra Token Plans are closed to new subscriptions. If your legacy subscription auto-renews, you can keep its existing terms or move to M Plan.
Read More: Generate Text, Image, Video, Speech, and Music via CLI – MiniMax CLI
Xiaomi MiMo Token Plan
Official plan: Xiaomi MiMo Token Plan
Xiaomi MiMo Token Plan has four monthly tiers: Lite at $6, Standard at $16, Pro at $50, and Max at $100. Monthly credit allowances are 4.1B, 11B, 38B, and 82B respectively. MiMo V2.6 Pro and V2.6 Flash can be used through coding clients configured for the plan. MiMo V2.5 and V2.5 Pro reached their scheduled retirement date on October 21, 2026, at 10:00 a.m. Beijing time. Their legacy model IDs have been retired.
MiMo Token Plan uses monthly credits and has no five-hour or weekly subscription quota. Requests from 00:00 to 08:00 Beijing time consume credits at 0.8x the regular rate. Xiaomi MiMo Desktop Membership has its own subscription and allowance.
Read More: MiMo Code: Free Open-Source AI Coding Agent with Persistent Memory
Multi-model AI coding subscriptions
Ollama charges cloud-model usage against credits. OpenCode Go meters usage by model over five-hour, weekly, and monthly periods.
Ollama Pay as you go, Pro, and Max
Official plans: Ollama Pay as you go, Pro, and Max pricing
Ollama Pay as you go has no monthly subscription fee and charges for cloud-model tokens using purchased credits. Running models locally is free. Pro costs $20 per month with $60 in cloud credits, and Max costs $100 with $300. These credit-based plans have no session or weekly usage caps. Pay as you go and Pro allow three concurrent requests, and Max allows ten.
Pro and Max credits reset on the monthly subscription anniversary and do not carry over. Once the plan balance is exhausted, purchased usage credits pay for further cloud requests. Requests over the concurrency limit are queued when capacity permits.
OpenCode Go and Go Plus
Official plan: OpenCode Go pricing and limits
OpenCode Go costs $10 per month. Go Plus costs $40 with higher per-model allowances. Both can be used with OpenCode and third-party coding agents. The model catalog includes GPT 6 Luna, Claude Haiku 5.5, GLM, Kimi, MiMo, MiniMax, Qwen, DeepSeek, and Grok models. Some models have regional or workspace privacy requirements.
OpenCode Go assigns a usage budget to each model. Five-hour and weekly allowances equal 20% and 50% of the model’s monthly dollar-value limit. On Go, GLM-5.3-Flash has a $60 monthly cap and Kimi K3.1 Pro has a $15 cap. Go Plus raises those limits to $180 and $60. DeepSeek V4.1 Flash has a $100 monthly cap on Go and $200 on Go Plus. When you reach a model’s subscription limit, you can use OpenCode Zen credits purchased through Zen billing for further requests.
AI coding plan usage limits and overage billing
Subscription login vs API key
A coding agent can charge the wrong billing account when an old API key or environment variable overrides its subscription credentials. Codex uses ChatGPT sign-in for subscription usage, and Claude Code uses a Claude account. xAI’s subscription draws Grok API requests and Cursor’s Grok requests from the weekly AI pool. GitHub Copilot and Devin bill through their product accounts. QwenCloud, MiniMax, and other API-based plans require the plan’s subscription key and endpoint.
After changing credentials, restart the coding agent and send one small request. Check the subscription usage page and the API billing page before starting a long session. The subscription meter should record the request if the intended plan is active. QwenCloud Token Plan keys work only with the Singapore-region subscription endpoint. Requests made with its regular API key or through another region do not deduct Token Plan credits.
What happens when the quota runs out?
QwenCloud stops requests at the monthly Credits limit unless you upgrade or use a Credit Pack. BytePlus also stops at the plan quota. MiniMax M Plan can use purchased credit packs after its subscription allowance is exhausted, but the Subscription Key never charges the pay-as-you-go API balance. xAI refreshes its Grok and Cursor AI pool weekly and sells top-ups. Claude and Codex offer paid continuation through the account’s usage or billing settings when credit purchases are enabled. GitHub Copilot, Devin, Ollama, Mistral, Google, Kimi, and OpenCode also offer paid overages or top-ups. Ollama’s local models continue running on your own hardware.
Check the fallback setting before a long coding session. A hard-stop plan prevents unplanned charges. A credit-based fallback continues requests after the subscription allowance is exhausted and can increase the monthly bill.
Can I share an AI coding plan?
Personal coding subscriptions generally restrict credentials to one subscriber. QwenCloud, GLM, Tencent Token Plan, Kimi Code, and similar plans restrict sharing. Some also prohibit automated or backend workloads. Team use requires a subscription with team seats or organization billing.
How to choose an AI coding plan
Free access varies by coding environment. GitHub Copilot Free and Devin Free support editor-based coding, and xAI Free provides limited Cursor and Grok Build access.
For IDE coding, compare Copilot’s monthly Credits, Devin’s daily and weekly agent limits, and xAI’s weekly Grok and Cursor pool. For terminal agents, confirm the subscription endpoint works with your client and check its five-hour, weekly, or monthly quota.
Keep optional paid fallback disabled during your first billing cycle and use your own usage history to decide whether a larger tier is worth the extra cost.
FAQs
Does changing coding agents reset the quota?
No. The allowance belongs to the subscription account or key. Switching coding agents does not reset a GLM, QwenCloud, MiniMax, Kimi, MiMo, or OpenCode Go quota.
Can one coding agent use several subscriptions?
Yes, when the agent accepts multiple providers. Configure each subscription key with the correct base URL and provider profile. Each plan charges its own billing account and uses its own quota.
Does OpenAI-compatible mean every plan works in every agent?
No. OpenAI-compatible describes the API request format. The agent must also accept a custom provider, use a model ID the plan recognizes, and connect to the correct subscription endpoint.
Does DeepSeek have a monthly Coding Plan?
DeepSeek does not currently offer a fixed monthly Coding Plan. Its first-party API uses per-token billing. Several subscriptions in this comparison include DeepSeek models, including OpenCode Go, BytePlus ModelArk, QwenCloud Token Plan, and Tencent Cloud Token Plan.
Read More: How to Use DeepSeek in Codex and Claude Code
Related resources
- Claude Code Pricing and Usage Limits: Free, Pro, Max, Ultra, Team, and API
- AI LLM API Pricing: GPT, Gemini, Claude & More
- 10 Best CLI AI Coding Agents: Open-Source & Commercial
- Gemini API Free Tier Limits: Requests Per Day & Setup
- OpenAI API Rate Limits: RPM, TPM, Tiers, and 429 Errors









