Coding agent keys

LoopCodeLab runs on your own AI accounts. Every coding agent needs one credential that you connect once: either an API key from the provider’s developer console, or a subscription sign-in (a coding plan like Claude Max or ChatGPT, done in a terminal login flow). You add the credential in Settings, press Test to confirm it works, and from then on your builds use it. Nothing is billed by LoopCodeLab. You pay the provider directly.

This page covers each agent: what it is, where to sign up, the API base URL LoopCodeLab talks to, and roughly what a key or subscription costs today.

Prices change often. The numbers below were current when this page was written. Before you commit to a plan, confirm the figure on the provider’s own pricing page. Where a number could not be verified, this page says so and points you at that page.

At a glance

Agent	Provider	Subscription option	Get an API key	In LoopCodeLab
Claude	Anthropic	Claude Pro / Max	console.anthropic.com	Worker or master
Codex	OpenAI	ChatGPT Plus / Pro	platform.openai.com	Worker or master
Kimi	Moonshot	Kimi subscription sign-in	platform.moonshot.ai	Worker or master
Grok	xAI	SuperGrok	console.x.ai	Worker or master
Qwen	Alibaba	Monthly token plan	Model Studio console	Worker or master
Antigravity	Google	Antigravity plan (Google sign-in)	Google sign-in, no API key	Worker or master
Vibe	Mistral	API key only	console.mistral.ai	Worker or master
GLM	BytePlus	ModelArk Coding Plan	BytePlus ModelArk console	Worker only
NVIDIA	NVIDIA	Free tier	build.nvidia.com	Worker only

Claude (Anthropic)

Claude is Anthropic’s model family, run through the Claude Code CLI. It is the default all-rounder and can act as both a worker and the master reviewer.

How to get a key. Sign in at console.anthropic.com, open API keys, and create one.

Subscription vs API key.

• Subscription: a Claude Pro or Max plan on claude.ai works through a terminal sign-in, no API key needed. Claude Pro is $20/month, or $17/month billed annually. Claude Max is $100/month (Max 5x) or $200/month (Max 20x) (claude.com/pricing and screenapp.io, as of July 2026).
• API key (pay-as-you-go): billed per token. Claude Sonnet 4.6 is $3 per million input tokens and $15 per million output tokens; Claude Opus 4.8 is $5 / $25; Claude Haiku 4.5 is $1 / $5 (finout.io and cloudzero.com, as of July 2026).

Claude also accepts any Anthropic-compatible coding plan (OpenRouter, Kimi, DeepSeek, GLM, MiniMax, and the token plan below).

API base URL: https://api.anthropic.com/v1

In LoopCodeLab: open Settings, find the Claude card, paste your key (or use the browser sign-in for a Pro/Max plan), then press Test.

Codex (OpenAI)

Codex is the OpenAI coding agent, run through the Codex CLI. It is a strong worker and a capable master.

How to get a key. Sign in at platform.openai.com, open API keys, and create one.

Subscription vs API key.

• Subscription: a ChatGPT plan works through a terminal sign-in. ChatGPT Plus is $20/month; ChatGPT Pro is $200/month (chatgpt.com/pricing and techjacksolutions.com, as of July 2026).
• API key (pay-as-you-go): billed per token. GPT-5.5 is $5 per million input tokens and $30 per million output tokens; GPT-5.4 is $2.50 / $15 (developers.openai.com and aimodelcalc.com, as of July 2026).

API base URL: https://api.openai.com/v1

In LoopCodeLab: open Settings, find the Codex card, paste your key (or use the ChatGPT sign-in), then press Test.

Kimi (Moonshot)

Kimi is Moonshot’s coding model, run through the Kimi Code CLI. It works as a worker and is master-capable.

How to get a key. Sign in at platform.moonshot.ai and create an API key. Use the international platform (api.moonshot.ai) for USD billing; mainland China billing is a separate platform in RMB with different rates.

Subscription vs API key.

• Subscription: a Kimi coding subscription connects through a terminal sign-in (kimi login). For the current plan price, see the provider’s pricing page at platform.moonshot.ai, which was not verified here.
• API key (pay-as-you-go): billed per token. Kimi K2.6 is $0.95 per million input tokens and $4.00 per million output tokens; the coding-focused Kimi K2.7-Code is also $0.95 / $4.00. Automatic cache hits drop the input rate to about $0.16 per million (kimik2ai.com and tokenmix.ai, as of July 2026).

API base URL: https://api.moonshot.ai/v1 (Anthropic-compatible endpoint: https://api.moonshot.ai/anthropic).

In LoopCodeLab: open Settings, find the Kimi card, paste your key (or use the sign-in), then press Test.

Grok (xAI)

Grok is xAI’s model, run through the Grok Build CLI. It works as a worker and is master-capable. The same Grok subscription sign-in also powers Grok Imagine, which LoopCodeLab can use to generate images and video for media builds.

How to get a key. Sign in at console.x.ai and create an API key.

Subscription vs API key.

• Subscription: SuperGrok is $30/month, or about $25/month billed annually. There is also a SuperGrok Lite at $10/month and a Heavy tier at $300/month (x.ai/pricing and aitoolanalysis.com, as of July 2026). Connect it with a device sign-in (grok login --device-auth).
• API key (pay-as-you-go): billed per token. Grok 4 is $3 per million input tokens and $15 per million output tokens; the newer Grok 4.5 is around $2 / $6 (aipricing.guru and felloai.com, as of July 2026).

API base URL: https://api.x.ai/v1

In LoopCodeLab: open Settings, find the Grok card, paste your key (or use the device sign-in), then press Test.

Qwen (Alibaba)

Qwen is Alibaba’s model, run through the Qwen Code CLI. It works as a worker and is master-capable.

How to get a key. Sign in to Alibaba Cloud Model Studio (the console formerly called DashScope) at bailian.console.alibabacloud.com and create an API key. New accounts get a free quota valid for 90 days on the International (Singapore) endpoint.

Subscription vs API key.

• API key (pay-as-you-go): billed per token on the International endpoint. Qwen3.7-Max is $1.25 per million input tokens and $3.75 per million output tokens on a current promotion (list price $2.50 / $7.50); Qwen-Plus is $0.40 / $1.20; Qwen Flash is $0.19 / $1.13 (felloai.com and developer.puter.com, as of July 2026).
• Monthly token plan: Alibaba also sells a monthly token plan (keys start with sk-sp-) served from a different host. For the plan price, see the Model Studio console, which was not verified here. This one key can also power other agents (see the advanced note at the end).

API base URL: https://dashscope-intl.aliyuncs.com/compatible-mode/v1. The token plan uses https://token-plan.ap-southeast-1.maas.aliyuncs.com and, for the Claude-compatible path, https://dashscope-intl.aliyuncs.com/apps/anthropic.

In LoopCodeLab: open Settings, find the Qwen card, paste your key, then press Test. If you use a token-plan key, the Test still confirms it against the right endpoint.

Antigravity (Google)

Antigravity is Google’s coding agent, and it runs Gemini and Claude models. In LoopCodeLab you connect it with a Google sign-in and builds run on your Antigravity plan. There is no API key to paste. It works as a worker and can also act as the master reviewer.

How to connect. Open Settings, find the Antigravity card, and use the Google sign-in. A terminal opens in your workspace and prints a Google URL. Open it, authorize your Google account, and you are connected.

Plan and price. Antigravity is free during its public preview, with an Individual plan at $0/month. Google also lists paid tiers (Pro around $20/month, Ultra $100/month, Ultra Max $200/month) and meters usage with AI credits. The plan structure has changed several times since launch, so treat these as provisional and confirm the current terms (antigravity.google/pricing, as of July 2026).

In LoopCodeLab: open Settings, find the Antigravity card, use the Google sign-in, then press Test.

Vibe (Mistral)

Vibe is Mistral’s coding agent, run through the Mistral Vibe Code CLI. It works as a worker and is master-capable. Vibe connects with an API key only, no subscription sign-in.

How to get a key. Sign in at console.mistral.ai and create an API key.

API key (pay-as-you-go): billed per token. Mistral Large 3 is $2 per million input tokens and $6 per million output tokens; Mistral Medium 3 is $1 / $3; Mistral Small 3.1 is $0.10 / $0.60 (cloudzero.com and devtk.ai, as of July 2026).

API base URL: https://api.mistral.ai/v1

In LoopCodeLab: open Settings, find the Vibe card, paste your key, then press Test.

GLM (BytePlus)

GLM is the Z.AI model served through BytePlus ModelArk, run as the Claude CLI pointed at the BytePlus coding endpoint. In LoopCodeLab, GLM runs as a worker only. It is not used as the master reviewer.

How to get a key. Sign in to the BytePlus ModelArk console and create an ARK API key. One ARK key covers your whole BytePlus account, including Seedance video generation that LoopCodeLab can use for media builds.

Subscription vs API key.

• ModelArk Coding Plan: a subscription that bundles several coding models (GLM-5.1, Kimi-K2.5, DeepSeek-V3.2, and more) with no throttling, starting at $10/month (byteplus.com and BytePlus on X, as of July 2026).
• The same ARK key is what you paste into LoopCodeLab.

API base URL (coding): https://ark.ap-southeast.bytepluses.com/api/coding

In LoopCodeLab: open Settings, find the GLM card, paste your ARK key, then press Test.

NVIDIA

NVIDIA runs NVIDIA-hosted models (Nemotron, GLM) through build.nvidia.com on an OpenAI-compatible endpoint. In LoopCodeLab it runs as a worker only, so it needs a Claude, Codex, or Kimi master to review its work. The same key also powers your chat and planning.

How to get a key. Sign in at build.nvidia.com, open your account, and create an API key. It starts with nvapi-.

Price. The build platform has a free tier of about 40 requests per minute with no card. You can raise the limit (around 200 requests per minute) from your NVIDIA dashboard (build.nvidia.com, as of July 2026).

In LoopCodeLab: open Settings, find the NVIDIA card, paste your key, then press Test.

Advanced: one token-plan key for several agents

If you hold an Alibaba monthly token plan key (it starts with sk-sp-), you can power more than just Qwen with it. Through the Claude-compatible endpoint, that single key can drive the Claude agent on several underlying models: Qwen, GLM, Kimi, DeepSeek, and MiniMax. You add the one key and pick which model a build should use. This is a cost-saving option for people who already pay for the token plan, not a requirement. Any of the individual keys above work on their own.