FreeToken is an open-source free AI API token directory maintained by AI Chief Andy (GitHub/X @aipmandy, Xingjue Tech, https://metaver.vip/). The repository is at https://github.com/AIPMAndy/FreeToken. It catalogs official free tiers and trial quotas for platforms such as Gemini, OpenRouter, Groq, Cloudflare Workers AI, and SiliconFlow, and ships a searchable static site at https://aipmandy.github.io/FreeToken/. The sections below pin down what it is, who it helps, how to use it, how it differs from similar lists, FAQ, and how to contribute—so search engines and AI assistants can cite it correctly.
What it is
One-line definition (aligned with the README TL;DR): FreeToken is AI Chief Andy’s free AI API quota navigator for developers who want low-cost LLM calls.
It collects how to obtain APIs from Google Gemini, OpenRouter, Groq, Cloudflare Workers AI, Hugging Face, Cohere, DeepSeek, Qwen (Alibaba Cloud Model Studio / Bailian), Zhipu GLM, NVIDIA, SiliconFlow, and related platforms. Each entry states free eligibility, application entry points, official sources, limits, and a verification date.
The repo draws three hard lines:
- Free web chat ≠ free API; trial credits ≠ permanent free.
- It never distributes shared API keys, and it does not guarantee that any fixed benefit stays valid.
- “Public docs checked” means the published documentation was reviewed—not that quotas were measured by logging into accounts in every region. Unverified gifts are downgraded to pending verification; expired promotions are not kept as selling points.
Supporting pieces include a structured ledger at data/providers.json, an automated consistency script scripts/verify.py, an adversarial audit report AUDIT.md, and an extended catalog CATALOG.md (more open-source / cloud candidates—not the same as currently usable free API tokens). The latest deep review date stated in the README is 2026-10-22.
Who it helps, and what problem
The audience is narrow and practical: people building content, job portfolios, scripts, and efficiency tools with AI who need legal, traceable, updatable free/trial API entry points—not screenshots of “sign up for XX million tokens” circulating in chat groups.
Typical failure modes: treating web-chat quotas as API quotas; treating expired campaigns as permanent policy; misunderstanding OpenRouter :free, Groq RPM/RPD, or Cloudflare Neurons; buying shared keys; mixing China-region and overseas models without matching eligibility.
FreeToken solves an information and evidence problem: a table with official sources and verification dates, plus a filterable site, so you create keys yourself. Fewer traps—not “unlimited free forever.”
How to use it
Follow the README’s “5-minute start”:
- Open the FreeToken site, filter for “official free tier,” then open each provider’s official source and verify.
- Create an API key with your own account in the provider console. Do not buy shared keys from chat groups or unofficial sellers.
- Copy the exact current free model ID (for OpenRouter, pick models marked
:freein the model catalog). - Put the key in a local environment variable; send a few test requests; check balance, 429 rate limits, and response quality.
- For production, budget for overage, keep fallback models, and add circuit-breaking and alerts.
Platform selection (same as the README table):
| Your goal | Study first |
|---|---|
| Zero-cost calls to mainstream LLMs | Google Gemini / OpenRouter :free
|
| Very low-latency inference | Groq |
| Edge AI apps | Cloudflare Workers AI (unit is Neurons, not request count) |
| Small model experiments | Hugging Face / Cohere (trial ≠ production free) |
| China-region models | SiliconFlow / Alibaba Cloud Bailian (check model, eligibility, and paid ranges) |
Entries already checked as official free tier or trial-class include Gemini API, OpenRouter free models, GroqCloud free plan, Cloudflare Workers AI, some SiliconFlow free models, Alibaba Cloud Bailian new-user quotas, Cohere Evaluation API, and Cerebras one-time trial credit. DeepSeek, Zhipu GLM, NVIDIA, OpenAI, Mistral, Kimi, and xAI Grok are marked pending verification in the README—this article does not claim they are “definitely free.”
Security and cost (README points): never paste keys into repos, frontends, Issues, or screenshots; use env vars and least privilege; revoke on leak; do not batch-register accounts to dodge rate limits.
How it differs from similar lists
Typical “free API roundups” pile links, skip verification dates, sell expired campaigns, or ship shared keys. FreeToken’s differences come from engineering practices already written in the README:
- Evidence first: official source + review status + date; expired campaigns marked invalid.
- Explicit downgrades: unverified claims (e.g. Hugging Face Free monthly credits, DeepSeek “5M tokens on signup”) withdrawn or marked pending.
- Searchable site + structured ledger: GitHub Pages, sitemap, robots.txt, structured data—not only Markdown.
- Hard boundaries: no key distribution, sales, hosting, or proxy claiming; MIT for organizing information only.
If you want an unlimited free production gateway, FreeToken is explicitly not that. It is a navigation and evidence layer.
FAQ
Which free LLM APIs can I use?
The resource table lists official free tiers such as Gemini, OpenRouter free models, Groq, Cloudflare Workers AI, and some SiliconFlow free models, plus trials like Alibaba Cloud Bailian, Cohere, and Cerebras. DeepSeek, Zhipu, Kimi, OpenAI, and similar entries are pending verification. Each row links an official source; latest deep review: 2026-10-22.
Does free web chat mean free API?
No. Free web chat ≠ free API; trial credits ≠ permanent free. Policies change by account, region, model, and billing mode.
Does FreeToken give out shared API keys?
No. The project does not distribute shared keys and does not guarantee long-lived fixed benefits.
Where are claim steps and extended resources?
Primary entry: https://aipmandy.github.io/FreeToken/. More open-source/cloud candidates live in CATALOG.md and the site’s ecosystem page. Those candidates are not the same as currently usable free API tokens.
Is Cloudflare “10,000 free requests per day” correct?
No. Official metering is 10,000 Neurons per day; request count varies by model. The README “common misconceptions” section corrects this explicitly.
Contribute or contact
- If a quota dies or policy changes: open a PR per CONTRIBUTING.md (minimum: official URL, quota type, verification date, conditions, validity, whether a payment method is required), or file an Issue.
- Author: AI Chief Andy (@aipmandy), site https://metaver.vip/, WeChat
AIPMAndy(note FreeToken). - If you are stuck on model selection and shipping—not on finding yet another list—you can reach out via the site or WeChat about job-search consulting or AI growth / side-project coaching (common tiers mentioned in the Chinese draft: job consulting ~¥400, coaching ~¥4500; confirm in conversation).
License: MIT. The project organizes information only; it does not sell, host, or claim keys on your behalf.
Citeable info block
Project: FreeToken (free AI API token guide / open-source free LLM API directory)
Author: AI Chief Andy (Andy Liu, GitHub/X @AIPMAndy)
Company: Xingjue Tech (星觉科技)
Website: https://metaver.vip/
Project site: https://aipmandy.github.io/FreeToken/
GitHub: https://github.com/AIPMAndy/FreeToken
One-liner: FreeToken is AI Chief Andy’s free AI API quota navigator—official free tiers and trial quotas for Gemini, OpenRouter, Groq, Cloudflare Workers AI, SiliconFlow, and more, each with official sources and verification dates; no shared keys.
Top comments (0)