AI Coding Plans Compared: Who You Contract With and Where Your Code Is Processed

An AI coding plan is a monthly subscription your coding agent draws on, with usage windows instead of a bill per token. Many plans offer the same open models, such as GLM or Kimi; what differs is the company you sign with and where your prompts, and the code in them, are processed. This comparison looks at five plans: Nodion AI Coding Plans, the Z.ai GLM Coding Plan, Kimi Code, the MiniMax M Plan and Synthetic. Among the plans compared, Nodion is the only one offered by a company based in the EU. As of October 2026; prices are left out, because they change.

Three things to keep apart

Privacy pages answer different questions, so read them separately:

• Who you contract with. The contracting company and its country decide which law and which courts apply.
• Where account data is stored. Privacy policies usually describe account, billing and usage data, not where the model runs.
• Where inference runs. This is where your prompts and your code are processed. Several providers do not state it for their plan.

Then there is retention and training. Most plans have no statement of their own, only general terms; the tables quote those and mark what is not stated.

At a glance

Plan	Contracting company	Account and personal data	Where inference runs
Nodion AI Coding Plans	Nodion GmbH, Germany	APIs and control plane hosted in Germany	EU: all models provided for the plans run in the EU
Z.ai GLM Coding Plan	Jingsheng Hengxing Technology Pte. Ltd., Singapore	Generally processed in Singapore	Not stated for the plan
Kimi Code (kimi.ai)	NOVASCENT PRIVATE LIMITED, Singapore	May be stored on servers outside your country of residence, no country named	Not stated
MiniMax M Plan	Nanonoble Pte. Ltd., Singapore	Stored in a data center in the United States	Not stated
Synthetic	Synthetic Lab, Co., USA	Processed in the United States	Not stated; inference partners not named

Plan	Retention and training (as stated)	Official tools	API
Nodion AI Coding Plans	Zero data retention for all models served through the Nodion AI Gateway	OpenCode, Codex CLI, Aider, Cline, Continue, any OpenAI-compatible client	OpenAI-compatible: Chat Completions and Responses
Z.ai GLM Coding Plan	Individual users: content may be used to improve services; API: not stored, not used for improvement without consent; Team plan: no training by default. Which covers the individual plan is not stated	Officially supported tools only, such as Claude Code, Codex, OpenCode, Cline, Cursor	Anthropic-compatible and OpenAI-style
Kimi Code (kimi.ai)	Content used to provide and improve the services, including training; opt-out by email	Kimi Code CLI, desktop, VS Code; guides for Claude Code, OpenCode, Codex	OpenAI- and Anthropic-compatible
MiniMax M Plan	Input and output may be used to develop and improve services; no retention period stated	MiniMax Code, Claude Code, Codex, Cursor, OpenCode and others	Anthropic- and OpenAI-compatible
Synthetic	API: prompts and completions not stored without consent (API only, not the web UI); no training on prompts or completions	Guides for Claude Code, OpenCode, Crush and others	OpenAI- and Anthropic-compatible

1. Nodion AI Coding Plans

AI Coding Plans come from Nodion GmbH, based in Stuttgart, Germany, in three sizes: Go, Pro and Max. A plan pays for chat models and for tools such as web search, news and fetch. The chat models run in the EU, at providers based in the EU, all serving with zero data retention. Usage counts against three windows: a 5-hour session, the week and the month. When a window is full, requests stop until it resets, unless the plan allows wallet use; then they continue from the prepaid wallet.

You use a plan key in OpenCode, Codex CLI, Aider, Cline, Continue and any OpenAI-compatible client, with the base URL https://api.nodion.ai/v1. Codex uses the Responses API (wire_api = "responses"). Claude Code and Cursor cannot call this API and keep their own models; they can still deploy to Nodion over MCP. The plan covers models such as Kimi K3, GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash and Qwen3.8 27B. For teams, the project owner buys one plan per developer and assigns it to a project member. Setup guides: Codex CLI, OpenCode, and Aider, Cline and Continue.

Sources: AI Coding Plans, AI Inference docs, Responses API, Regions, Legal notice.

2. Z.ai GLM Coding Plan

Z.ai's GLM Coding Plan comes in Lite, Pro and Max tiers for individuals, plus a Team plan, with GLM-5.3 and GLM-5.3-Flash. The contracting company is Jingsheng Hengxing Technology Pte. Ltd. in Singapore, under Singapore law. The privacy policy says personal data is generally processed in Singapore and may be transferred to group companies or third parties overseas. Where the plan's inference runs is not stated separately; the data processing terms for Z.ai's API services also say customer data is generally processed in Singapore.

On training, Z.ai's documents have two regimes. For individual users, the terms of use reserve the right to process user content to improve the services and to develop new products. For API services, the data processing terms say content is not stored, and the terms of use say it is not used to develop or improve the services unless you explicitly agree. Which of the two covers the individual Coding Plan is not stated; the Team plan page advertises that code and prompts are excluded from training by default. The plan may only be used in officially supported tools, not for general API access or your own apps. Usage has 5-hour and weekly limits.

Sources, checked October 2026: overview, privacy policy and DPA, terms of use, subscription terms, Team plan, tools, FAQ.

3. Kimi Code

Kimi Code is the coding service included in Moonshot AI's Kimi membership, from the Plus tier up. For international users on kimi.ai, the contract is with NOVASCENT PRIVATE LIMITED in Singapore, under Singapore law. The kimi.ai privacy policy says your information may be transferred to and stored on servers outside your country of residence, without naming a country; it does not say where inference runs. Moonshot's site for China, kimi.com, has its own company and privacy policy; this comparison cites kimi.ai.

The kimi.ai privacy policy says user content such as prompts and files is processed to provide and improve the services, including training its AI models. The terms let you opt out of training by email; the opt-out applies only going forward. The policy states no fixed retention period. Usage runs on a rolling 5-hour window and a monthly total shared with the membership.

Sources: Kimi Code and membership, kimi.ai terms, kimi.ai privacy policy, Kimi Code docs, limits.

4. MiniMax M Plan

The M Plan is MiniMax's subscription for individuals, in the tiers Go, Explore and Build; it follows the earlier Coding Plan and Token Plan and covers text, video, image and audio models. On the international platform the operator is Nanonoble Pte. Ltd. in Singapore, under Singapore law. The privacy policy says personal data is stored in a data center in the United States and transferred to countries or regions outside your country of residence. Where inference runs is not stated; the US statement is about storage.

The terms say MiniMax may use input and generated content to provide, maintain, develop and improve its services. The privacy policy adds that personal data is not used for training to profile or target consumers; it states no retention period for prompts. Usage runs on a 5-hour window and a weekly window.

Sources: M Plan, Token Plan notice, terms of service, privacy policy, other tools, usage rules.

5. Synthetic

Synthetic is run by Synthetic Lab, Co. in San Francisco, with Go, Standard and Big subscriptions for open models such as Kimi-K3, GLM-5.3-Flash and DeepSeek-V4.1-Flash. Its privacy policy says personal information is processed in the United States; which inference partners run the models, and in which countries, is not stated.

For API use, the policy says prompts and completions are not stored without your explicit consent and are deleted once completed; this applies to the API, not to use through Synthetic's web interface. Synthetic says it does not use prompts or completions to train or improve models, and works only with inference partners that also do not store them or train on them. Credits regenerate over the week.

Sources: plans and models, privacy policy, API, guides.

How to choose

• Models that run in the EU, under a contract with an EU company: Nodion AI Coding Plans; among the plans compared, Nodion is the only one offered by a company based in the EU.
• Zero data retention for every model: Nodion serves all models through its gateway with zero data retention.
• Whichever plan you pick: read the terms that cover the plan itself, not only the general privacy policy, and check the three points above one by one. More in what zero data retention covers.