AI DEAL COLLECTION
AI deals that are easier to claim or use from China, with direct access or China-friendly providers.
AI deals that are easier to claim or use from China, with direct access or China-friendly providers. It is useful for developers, indie hackers, and AI tool users who want to compare free credits, limits, and alternative routes quickly.
yangmao.ai refreshes free tiers, expiration dates, claim requirements, and accessibility signals through automated pipelines plus manual checks. Always verify the final claim page before use.
Check the same page for alternative providers, OpenAI-compatible APIs, China-friendly access, or evergreen free tiers instead of relying on one vendor.
DeepSeek has extended its 2.5 discount promotion for the V4 Pro model until May 31, 2026. The offer applies to all API calls, providing users with significantly discounted access to the high-performance model. This is an excellent opportunity for developers to experience a powerful model at a low cost.
DeepSeek V4 Pro has an official limited-time 75% API discount through 2026-05-31 15:59 UTC. This is a pricing promo, not a verified free-credit grant.
新注册用户赠送500万token免费额度，支持 DeepSeek V3 模型，国内直接使用，无需翻墙。
新注册用户赠送 €10 API 额度，可用于 Mistral Large 等模型，支持国内邮箱注册，需绑定国际信用卡。
新注册用户可获 $5 API 额度，用于体验 o3-mini 模型，有效期30天，支持国内信用卡注册。
ActivePieces offers a free or open-source option: Open-source version free for self-hosting; cloud version offers free tier. Useful for low-cost developer testing.
LiveKit Agents offers a free or open-source option: Open-source framework is free; LiveKit cloud offers 1,000 free minutes/month. Useful for low-cost developer testing.
Free AI learning resource on yangmao.ai 原创: AI Coding — From Zero to Freelancing. Good for structured learning.
Free AI learning resource on Coursera: AI For Everyone. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: AI-Powered Graduate Exam English Prep. Good for structured learning.
Free AI learning resource on 多平台整合: AI Monetization Guide — 10 Ways to Make Money with AI. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: AI Presentations — Create Slides in 10 Minutes. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: AI Resume & Interview Prep for Students. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: AI Side Hustle Guide for College Students. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: AI Thesis Writing Guide — From Topic to Final Draft. Good for structured learning.
Aider offers a free or open-source option: Fully open-source and free, bring your own API key with no restrictions. Useful for low-cost developer testing.
AutoDL has recorded free compute or trial credits: ¥10 credits. Useful for inference, deployment, or GPU experiments.
AutoGen offers a free or open-source option: Open-source project, self-hostable, no usage limits. Useful for low-cost developer testing.
Awesome n8n Templates offers a free or open-source option: Completely free, open-source template collection on GitHub. Useful for low-cost developer testing.
百川智能为新注册用户提供 100 万 token 免费 API 额度，支持 Baichuan4 系列模型，国内直连，无需科学上网。
百川智能为 Baichuan4 模型提供新用户注册即送100万token免费API额度，支持中文优化，国内直接访问，适合开发者快速集成。
Baichuan AI is recorded as China-friendly: Chinese platform, direct access.. Useful when you need access without complex network setup.
注册百川智能开放平台即送 100 万 token，支持 Baichuan4 和 Baichuan3-Turbo 模型，国内直连，无需海外支付方式。
Baichuan AI API has a recorded free trial: 500万 tokens; rate limit: 5 RPM.
Baichuan AI has a recorded free tier: No explicit limit. Good for testing before upgrading.
Baichuan AI is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
Baichuan AI is recorded as supporting OpenAI-compatible API access. Free/trial info: 500万 tokens. Useful for low-cost testing by swapping SDK base_url.
百度千帆大模型平台为新用户提供100万Token免费调用额度（支持ERNIE 4.0、ERNIE Speed等），另赠50元体验金。国内开发者可直接使用百度账号注册，API兼容OpenAI格式，迁移成本低。
百度千帆大模型平台为新用户提供 100 万 token 的免费调用额度，支持 ERNIE-Bot、ERNIE-Bot-turbo 等模型，国内直接访问，注册即用，无需绑定支付方式。
百度千帆平台为新用户提供 ERNIE-Bot 系列模型免费调用额度，包含 100 万 tokens，支持 API 调用，国内直接可用，无需海外支付方式。
百度千帆平台近期调整免费政策，ERNIE-Bot、ERNIE-Bot-Turbo 等模型每日免费调用次数提升至 1000 次，注册即享，无需绑定银行卡，国内开发者友好。
百度千帆大模型平台为新用户提供 200万 token 免费额度，支持 ERNIE-Bot、ERNIE-Bot-turbo 等模型，国内网络直接使用，注册即送。
百度千帆大模型平台为新用户提供100万 token 免费额度，适用于 ERNIE 3.5 和 ERNIE 4.0 模型，支持文本生成、对话等场景。国内直接访问，无需科学上网，注册即用。
BeeLlama v0.2.0 brings a major DFlash update, dramatically improving single-GPU inference performance. On a single RTX 3090, Qwen 3.6 27B achieves 164 tps (4.40x improvement) and Gemma 4 31B reaches 177.8 tps (4.93x improvement). Prompt processing speed remains near baseline. This open-source tool is free to use for local deployment and efficient inference.
Bolt.new offers a free or open-source option: Free plan with limited daily tokens for generating and deploying apps. Useful for low-cost developer testing.
Free AI learning resource on 多平台整合: Build Your First LLM App with Free APIs. Good for structured learning.
Free AI learning resource on DeepLearning.AI: ChatGPT Prompt Engineering for Developers. Good for structured learning.
A developer built a free local MCP server that significantly optimizes Claude Code's PR review process. The tool reduces token consumption per PR review from 63K to 8.7K, drastically lowering usage costs. Users need to set up the local server and integrate it into their Claude Code workflow. This solution is ideal for developers who frequently use Claude Code for code reviews.
Claudex is a free open-source CLI tool built by a community developer, designed to emulate Claude Code-style workflows. Users can try it without any subscription fee, requiring only a Claude API key to run. It is suitable for developers exploring Claude's coding assistance capabilities and supports customizable workflows.
Cloudflare Workers AI is recorded as China-friendly: Direct access from China via Cloudflare edge network, low latency. Workers AI accelerated by global CDN.. Useful when you need access without complex network setup.
Cloudflare Workers AI API has a recorded free trial: 每天 10000 神经元（永久有效）; rate limit: 10000 requests/day.
Cloudflare Workers AI has a recorded free tier: 10,000 free requests/day. Good for testing before upgrading.
Cloudflare Workers $5/mo plan includes Workers AI with 10,000 free AI calls per day (measured in neurons), permanently valid. 50+ open-source models: - LLM: Llama 3.1 8B, Llama 3.3 70B, Gemma, Mistral 7B, Phi-2 - Image generation: Stable Diffusion XL (completely free!) - Embeddings: BGE Base/Large (for RAG and semantic search) - Speech-to-text: Whisper Highlights: - Permanently valid, never expires - Inference on 300+ global edge nodes, ultra-low latency - Direct China access, no proxy needed - OpenAI-compatible via AI Gateway - Pay-as-you-go after free quota, no hard cutoff - If you already use Cloudflare Workers, this is essentially free Ideal for lightweight AI: blog writing, content tagging, summarization, embeddings, product image generation.
Cloudflare Workers AI is recorded as supporting OpenAI-compatible API access. Free/trial info: 每天 10000 神经元（永久有效）. Useful for low-cost testing by swapping SDK base_url.
Code Relay is attractive because it may expose GPT-5.5 / Claude 4.7-style models through one relay, but it is still a third-party relay rather than an official provider. Security and long-term stability are materially weaker than official platforms. Treat it as a risk case or tiny test channel, not a production API.
Codeium offers a free or open-source option: Individual plan free forever with unlimited completions + AI chat, no credit card required. Useful for low-cost developer testing.
Context7 offers a free or open-source option: Open-source, self-hostable or direct use, no limits. Useful for low-cost developer testing.
Continue offers a free or open-source option: Fully open-source and free, bring your own API key or use local models. Useful for low-cost developer testing.
Coze (ByteDance) is recorded as China-friendly: By ByteDance. China version coze.cn direct access. International coze.com needs proxy.. Useful when you need access without complex network setup.
Coze (ByteDance) API has a recorded free trial: Free tier; rate limit: Varies.
Coze offers a free or open-source option: China version (coze.cn) basic features free, international version (coze.com) has free quota. Useful for low-cost developer testing.
Coze (ByteDance) has a recorded free tier: No explicit limit. Good for testing before upgrading.
CrewAI offers a free or open-source option: Open-source framework, free to integrate, no commercial restrictions. Useful for low-cost developer testing.
Cursor is recorded as China-friendly: Cursor IDE accessible from China. AI features routed through Cursor servers.. Useful when you need access without complex network setup.
Cursor offers a free or open-source option: Free plan includes 2000 completions + 50 premium requests (GPT-4/Claude) per month. Useful for low-cost developer testing.
Cursor has a recorded free tier: 2000 completions + 50 premium requests/mo. Good for testing before upgrading.
DeepSeek's official docs confirm a capacity expansion request path for API accounts that need higher concurrency than the default limits. DeepSeek matches appropriate concurrency based on submitted business needs, with no additional cost for capacity expansion. This is for teams or businesses needing higher DeepSeek V4 Pro / V4 Flash concurrency; it is not free token credit and is not automatic access.
DeepSeek 为新注册用户提供 500 万 token 免费 API 额度（含对话和代码模型），支持国内直接访问，无需海外信用卡。
DeepSeek is recorded as China-friendly: Chinese service, direct access in China, very fast, no proxy needed. Useful when you need access without complex network setup.
注册即送 500 万 token，支持 DeepSeek-V2 和 DeepSeek-Coder 模型，兼容 OpenAI API 格式，国内直连可用，无信用卡要求。
DeepSeek 为新注册用户提供 500 万 token 的免费 API 额度（含输入和输出），支持 DeepSeek-V2 等模型，国内可直接访问，无需海外信用卡。
DeepSeek API has a recorded free trial: $5; rate limit: 2 RPM.
DeepSeek 为新注册用户提供 500 万免费 tokens，支持 DeepSeek-V2 和 DeepSeek-Coder 模型，API 兼容 OpenAI 格式，国内可直接访问，无需海外信用卡。
DeepSeek 为新注册用户提供 500万 token 的免费 API 调用额度，支持 DeepSeek-V2 和 DeepSeek-Coder 模型，国内可直接访问，无需海外信用卡。
DeepSeek offers 50 free inferences daily (V3 + R1 models) plus $5 API credits on signup. R1 reasoning model excels at math and code, one of the best free AI options available.
DeepSeek has a recorded free tier: 50 requests/day. Good for testing before upgrading.
DeepSeek 为新注册用户提供 500 万 token 的免费额度（含输入和输出），可用于 DeepSeek-V3 和 DeepSeek-R1 模型 API，有效期 30 天，支持国内直接访问，无需翻墙。
DeepSeek 为新注册用户提供 500 万 Token 免费额度，可用于 DeepSeek-V2 和 DeepSeek-Coder 系列模型 API 调用，支持文本生成与代码补全，国内直接访问，无需翻墙。
DeepSeek 为新注册用户提供500万Token免费额度，可用于其最新大模型API调用，支持文本生成、代码编写等，国内可直接访问注册，无需海外信用卡。
Free AI learning resource on DeepSeek 官网: DeepSeek Official Tutorial. Good for structured learning.
DeepSeek is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
DeepSeek is recorded as supporting OpenAI-compatible API access. Free/trial info: $5. Useful for low-cost testing by swapping SDK base_url.
新注册 DeepSeek 平台即赠送 500 万 token 免费额度，可用于调用 DeepSeek-V2 等模型 API，支持国内网络直接使用，无需海外信用卡。
DeepSeek V4 Flash model has been released, but the community noticed the lack of local deployment options like Ollama, sparking heated discussion on Reddit. Users are calling for a non-cloud version to run the model on local hardware. No free tier or deployment timeline is available yet, but interest is high.
A developer trained a prompt injection detector using ml-intern and DeepSeek V4 Flash, which runs directly in the browser. This provides users with a lightweight, privacy-friendly way to detect prompt injection attacks. DeepSeek V4 Flash demonstrates its potential in security applications as the base model.
DeepSeek-V4 is officially released with a million-token context window, greatly enhancing long-text processing capabilities. The model is optimized for agent applications, supporting more complex multi-step reasoning and tool calling. Developers can use it for free via the API with no additional cost. It is one of the longest-context open-source models available, suitable for document analysis, codebase understanding, and more.
The DeepSeek V4 Pro pricing transition is scheduled around May 31, 2026. The current 25%-of-list promotional window moves to the announced 1/4-of-list pricing basis, so users should verify the exact input/output rates in the DeepSeek console. Plan usage costs before high-volume API workloads.
Dify offers a free or open-source option: Open-source version fully free to self-host; cloud version with 200 GPT-4 calls per month. Useful for low-cost developer testing.
Domestic open-source large language model downloads have exceeded 10 billion, marking the vigorous growth of China's open-source AI ecosystem. These models include open-source versions released by multiple well-known vendors and institutions, covering different parameter scales from lightweight to large. Users can download model weights for free for academic research, commercial applications, or further development. This milestone reflects the widespread recognition and adoption of domestic AI technology by the open-source community.
Doubao (ByteDance) is recorded as China-friendly: Chinese platform by ByteDance, direct access, fast. Servers across China.. Useful when you need access without complex network setup.
Doubao (ByteDance) API has a recorded free trial: 50万 tokens; rate limit: 5 RPM.
Doubao (ByteDance) has a recorded free tier: No explicit limit. Good for testing before upgrading.
Doubao (ByteDance) is recorded as supporting OpenAI-compatible API access. Free/trial info: 50万 tokens. Useful for low-cost testing by swapping SDK base_url.
Easy-Dataset offers a free or open-source option: Open source and free to use, no quota limits. Useful for low-cost developer testing.
ERNIE Bot (Baidu) is recorded as China-friendly: By Baidu, direct access in China, fast and enterprise-stable.. Useful when you need access without complex network setup.
ERNIE Bot (Baidu) API has a recorded free trial: Free tier; rate limit: 5 RPM.
ERNIE Bot (Baidu) has a recorded free tier: No explicit limit. Good for testing before upgrading.
Fireworks AI 提供每日 100 万 token 免费额度，支持 Llama 3、Mixtral、Gemma 等主流开源模型。API 兼容 OpenAI 格式，国内可直连，适合原型开发和轻量应用。
提供高速推理 API，支持 Llama、Qwen 等开源模型。新用户有每日免费的 token 额度，适用于开发和测试。
FreeLLMAPI offers a free or open-source option: Fully open-source, aggregates free tiers from 14 providers, ~1.3B tokens/month. Useful for low-cost developer testing.
FreeModel is the lower-friction option in this GPT-5.5 free trial batch: no card, quick signup, and useful for light testing. The key unknowns are whether the model is truly native GPT-5.5, whether the weekly quota resets reliably, and long-term service stability.
A practical free path for running open-source AI projects locally before renting servers or GPUs. Start on your own computer, get the environment and scripts working, then decide whether to move to the cloud.
A community user created a custom benchmark called HalBench to evaluate sycophancy and hallucination tendencies in AI models. The test covered four frontier models: Sonnet 4.6, Grok 4.3, GPT 5.4, and Gemini 3.1 Pro. Results show Gemini 3.1 Pro performing strongly across multiple metrics, providing valuable reference for developers selecting reliable models.
Gemma 4 offers a free or open-source option: Fully open-source, Apache 2.0 license, commercial use allowed. Useful for low-cost developer testing.
GitHub Copilot is recorded as China-friendly: Direct access from China. Supports VS Code and JetBrains.. Useful when you need access without complex network setup.
GitHub Education is one of the most reliable education-benefit paths for AI coding. Verified students and teachers can receive Copilot-related benefits for long-term learning and development.
GitHub Copilot offers a free or open-source option: Free plan with 2000 completions + 50 chats per month. Free for students and OSS maintainers. Useful for low-cost developer testing.
GitHub Copilot Free is the official free tier for AI coding tools: 2,000 completions and 50 agent/chat requests per month, with no credit card required according to GitHub’s pricing page.
GLHF.chat 提供 Llama、Mistral 等开源模型的免费 GPU 推理服务，注册即送每月 25 美元额度，无需绑定信用卡。支持国内网络访问，适合低成本运行大模型。
Google Antigravity still has official free weekly limits and Pro/Ultra higher-limit signals, but the Ultra USD $100 bonus-credit sub-offer expired on 2026-05-25 and is no longer presented as claimable.
Google AI (Gemini) is recorded as China-friendly: Partially accessible, VPN recommended for stability. Useful when you need access without complex network setup.
Google Cloud's official $300 free credit offer for new customers can support AI API and cloud POC workflows. Eligibility and regional availability should be checked in the Google Cloud signup flow.
Google AI (Gemini) API has a recorded free trial: 免费 API 无需信用卡; rate limit: 15 RPM (Flash).
Google AI (Gemini) has a recorded free tier: Gemini free tier unlimited. Good for testing before upgrading.
The official Gemini API / AI Studio no-card free tier now has an additional entry: beyond Gemini API Free Tier input/output tokens, Google's I/O 2026 Blog confirms that new AI Studio builders can deploy their first two apps to Google Cloud at no cost with no credit card required. Production use, higher limits, or projects with billing already enabled still follow Cloud Run / Paid Tier rules.
Google Gemini API 提供免费层级，每分钟最多60次请求，支持 Gemini 1.5 Flash 和 Gemini 1.5 Pro 模型，国内开发者可通过代理或直接访问（部分地区可用）。无需绑定信用卡即可开始使用。
Free AI learning resource on Google: Google Machine Learning Crash Course. Good for structured learning.
Gorilla has recorded free compute or trial credits: unlimited. Useful for inference, deployment, or GPU experiments.
Groq 提供每日100万Token免费API调用额度，基于其自研LPU芯片实现极速推理（支持Llama 3、Mixtral等模型）。注册需海外邮箱，但API国内可直连，适合低延迟场景。
Groq 提供基于 LPU 推理引擎的免费 API，支持 Llama 3、Mixtral 等模型，每天最多 1440 次请求，国内可直连，适合低延迟推理测试。
Groq 将免费套餐的每日 API 请求上限从 500 次提升至 1000 次，支持 Llama 3、Mixtral 等开源模型，国内开发者可直接通过 API 调用，无需绑定信用卡。
Groq uses proprietary LPU (Language Processing Unit) chips for the world's fastest AI inference. Free tier requires no credit card. Free tier details: - Llama 3.3 70B: 30 RPM, 6000 tokens/min, 14400 requests/day - Llama 3.1 8B: 30 RPM, 20000 tokens/min - Gemma 2 9B: 30 RPM, 15000 tokens/min - Mixtral 8x7B: 30 RPM, 5000 tokens/min - Llama 4 Scout/Maverick (newly added) Why Groq is so fast: - Custom LPU chip designed specifically for LLM inference - Deterministic execution, no GPU memory bandwidth bottleneck - Llama 3.3 70B output at 300+ tokens/s (GPU typically 30-50 tokens/s) - Ultra-low time-to-first-token, ideal for real-time chat and streaming Best for: - Real-time AI chat (speed is the core experience) - Agent tool calls (low latency = faster multi-step reasoning) - Streaming output (buttery smooth typewriter effect) - Rapid prototyping China accessible. OpenAI-compatible API, base_url is https://api.groq.com/openai/v1.
Groq 于2026年4月底上线Mixtral 8x7B免费推理服务，每日500次请求，无需信用卡，API兼容OpenAI格式，国内开发者可直接调用。
Groq 提供基于 LPU 的高速推理服务，Mixtral 8x7B 模型每日免费额度高达100万token，注册即用，国内可直接访问 API。
Haystack offers a free or open-source option: Open-source project, self-hostable, no usage limits. Useful for low-cost developer testing.
HuggingFace AI Agents Course offers a free or open-source option: Completely free, certificate on completion. Useful for low-cost developer testing.
HuggingFace LLM Course offers a free or open-source option: Completely free, uses HuggingFace ecosystem tools. Useful for low-cost developer testing.
HuggingFace MCP Course offers a free or open-source option: Completely free, certificate on completion. Useful for low-cost developer testing.
HFViewer is recorded as China-friendly: The site is directly accessible; model metadata depends on Hugging Face and may require a proxy in some regions.. Useful when you need access without complex network setup.
HFViewer has a recorded free tier: Free web access. Good for testing before upgrading.
Hugging Face is recorded as China-friendly: Accessible from China (proxy may be needed in some regions). hf-mirror.com is a Chinese mirror.. Useful when you need access without complex network setup.
Hugging Face API has a recorded free trial: Free tier; rate limit: Varies.
Hugging Face has a recorded free tier: Varies by model. Good for testing before upgrading.
Hugging Face 提供 Inference API 免费套餐，每月 3 万次调用，支持数千个开源模型（文本、图像、音频等），国内可访问但速度较慢，适合学习和实验。
Hugging Face 提供免费推理 API，可调用数千个社区模型（包括文本、图像、音频等），国内可直接访问，无需付费。
Hugging Face is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
Tencent Hunyuan is recorded as China-friendly: By Tencent, direct access in China, enterprise-stable.. Useful when you need access without complex network setup.
Tencent Hunyuan API has a recorded free trial: 100万 tokens; rate limit: 5 RPM.
Tencent Hunyuan has a recorded free tier: No explicit limit. Good for testing before upgrading.
Tencent Hunyuan is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
Tencent Hunyuan is recorded as supporting OpenAI-compatible API access. Free/trial info: 100万 tokens. Useful for low-cost testing by swapping SDK base_url.
Kimi (Moonshot AI) is recorded as China-friendly: Chinese service, direct access. Web version free and unlimited, one of the most popular AI assistants in China.. Useful when you need access without complex network setup.
Kimi (Moonshot AI) API has a recorded free trial: ¥15 + 充 $5 送 $5; rate limit: 3 RPM.
月之暗面（Moonshot AI）为 Kimi 大模型 API 新用户提供100万 token 免费额度，支持长上下文（128K），国内直接访问，无需代理。注册即送，可用于对话、文档分析等场景。
Kimi (Moonshot AI) has a recorded free tier: No explicit limit. Good for testing before upgrading.
Kimi (Moonshot AI) is recorded as supporting OpenAI-compatible API access. Free/trial info: ¥15 + 充 $5 送 $5. Useful for low-cost testing by swapping SDK base_url.
Kiro offers a free or open-source option: Free plan with limited AI interactions per month. Useful for low-cost developer testing.
Official Kiro signup credit for new users: get up to $20 toward your first paid-plan upgrade. Kiro Pro includes premium models such as Claude Opus 4.7, Opus 4.6, and Sonnet 4.6. A card is required, so use a secondary/virtual card and confirm or cancel auto-renewal after activation.
Kling AI (Kuaishou) is recorded as China-friendly: By Kuaishou, direct access in China, fast. Leading video generation quality domestically.. Useful when you need access without complex network setup.
Kling AI (Kuaishou) has a recorded free tier: 66 credits/day. Good for testing before upgrading.
Lambda Cloud has recorded free compute or trial credits: 无免费额度，但价格有竞争力. Useful for inference, deployment, or GPU experiments.
LangChain4j offers a free or open-source option: Open source and free under MIT license. Useful for low-cost developer testing.
DGX Cloud Lepton (formerly Lepton AI) is recorded as China-friendly: Founded by Chinese-American team, good China access. API is directly accessible.. Useful when you need access without complex network setup.
DGX Cloud Lepton (formerly Lepton AI) API has a recorded free trial: $10 free credits; rate limit: 10 RPM.
DGX Cloud Lepton (formerly Lepton AI) has a recorded free tier: 10M tokens/day. Good for testing before upgrading.
DGX Cloud Lepton (formerly Lepton AI) is recorded as supporting OpenAI-compatible API access. Free/trial info: $10 free credits. Useful for low-cost testing by swapping SDK base_url.
Free AI learning resource on YouTube / B站: Hung-yi Lee Machine Learning 2024. Good for structured learning.
LM Studio is recorded as China-friendly: Runs locally, no network needed. May need proxy to download software and models.. Useful when you need access without complex network setup.
LM Studio API has a recorded free trial: Unlimited; rate limit: Local.
LM Studio has a recorded free tier: Unlimited (runs locally). Good for testing before upgrading.
LM Studio is recorded as supporting OpenAI-compatible API access. Free/trial info: Unlimited. Useful for low-cost testing by swapping SDK base_url.
Lovable offers a free or open-source option: Free plan with 5 projects + limited messages per month. Useful for low-cost developer testing.
Free AI learning resource on yangmao.ai 原创: Master ChatGPT Prompts — From Beginner to Expert. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: Master Claude Prompts — XML Tags & Extended Thinking. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: Master DeepSeek Prompts — R1 Thinking & Chinese Best Practices. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: Master Gemini Prompts — Multimodal & Real-time Search. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: Master Midjourney Prompts — V7 Formula & Style Guide. Good for structured learning.
Free AI learning resource on yangmao.ai 原创: Master SD & FLUX Prompts — Free AI Image Generation Guide. Good for structured learning.
Mastra offers a free or open-source option: Open-source project, self-hostable, no usage limits. Useful for low-cost developer testing.
mcp-go offers a free or open-source option: Open source project, completely free to use. Useful for low-cost developer testing.
Free AI learning resource on Microsoft Learn: Microsoft Azure AI Fundamentals (AI-900). Good for structured learning.
Million Engine is recorded as China-friendly: Direct China access, optimized for Chinese developers, low latency. Useful when you need access without complex network setup.
Million Engine is recorded as supporting OpenAI-compatible API access. Free/trial info: 按量付费. Useful for low-cost testing by swapping SDK base_url.
MiniMax is recorded as China-friendly: Chinese platform, direct access, fast.. Useful when you need access without complex network setup.
MiniMax API has a recorded free trial: ¥15; rate limit: Varies.
MiniMax has a recorded free tier: No explicit limit. Good for testing before upgrading.
MiniMax is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
MiniMax is recorded as supporting OpenAI-compatible API access. Free/trial info: ¥15. Useful for low-cost testing by swapping SDK base_url.
Mistral AI 于2026年4月更新免费政策，Le Chat 平台每月提供100万token免费额度，支持Mistral Large 2模型，国内可直连。
Mistral AI 的 Le Chat 聊天应用提供免费无限对话，支持 Mistral Large 等模型，国内可直接访问网页版，无需注册即可使用基础功能。
Mistral AI 的 Le Chat 平台提供免费层，支持无限次对话、文件上传（图像、PDF、Word、Excel）和网络搜索，无需付费。国内可直接访问网页版。
Mistral AI 推出的 Le Chat 聊天助手提供每日100次免费对话额度，使用自家 Mistral Large 模型，支持中文。可通过网页或 API 使用，注册即享，无需付费。国内可正常访问。
注册月之暗面开放平台即送 1500 万 token，支持 Kimi 长上下文模型（128K），国内直连，适合长文本处理任务。
月之暗面（Moonshot AI）为新注册用户提供 100 万免费 tokens，支持长上下文模型，API 兼容 OpenAI 格式，国内直接使用。
月之暗面 Moonshot 为新注册用户提供 150万 token 的免费 API 额度，支持 Moonshot-v1 模型，国内可直接访问，适合长文本处理。
月之暗面 Kimi 大模型为新注册开发者提供 500 万 token 的免费 API 调用额度，支持长上下文模型，国内网络可直接使用，适合构建对话和文本处理应用。
n8n offers a free or open-source option: Open-source version fully free to self-host; cloud version free with 5 workflows. Useful for low-cost developer testing.
Notion AI is recorded as China-friendly: Notion accessible from China (occasionally unstable in some regions).. Useful when you need access without complex network setup.
Notion AI has a recorded free tier: Limited requests. Good for testing before upgrading.
NVIDIA Build (NIM API) is recorded as China-friendly: Direct access from China via integrate.api.nvidia.com, no proxy needed. Medium speed, may slow during peak hours.. Useful when you need access without complex network setup.
NVIDIA Build (NIM API) API has a recorded free trial: 无限制（已取消额度限制）; rate limit: 40 RPM（可申请提升到 200 RPM）.
NVIDIA Build (NIM API) has a recorded free tier: Unlimited (40 RPM rate limit). Good for testing before upgrading.
NVIDIA Build (NIM API) is recorded as supporting OpenAI-compatible API access. Free/trial info: 无限制（已取消额度限制）. Useful for low-cost testing by swapping SDK base_url.
NVIDIA NIM has recorded free compute or trial credits: 40 RPM (upgradable to 200). Useful for inference, deployment, or GPU experiments.
NVIDIA offers free inference API for 100+ AI models spanning LLM, vision, speech, biology, and simulation. No credit card, no token limits, 40 RPM only. China accessible. 2026 highlight models: - Kimi K2.5 (Moonshot flagship, 1M context) - GLM-5.1 (Zhipu AI latest flagship, GLM-5 deprecated 04/20) - MiniMax M2.7 (230B params, coding/reasoning SOTA) - DeepSeek V3.2 / R1 (671B MoE, top reasoning) - Qwen 3.5 (397B/17B active, native multimodal) - Nemotron-3-Super-120B (NVIDIA's own, 1M context, Hybrid Mamba-Transformer) - Llama 4 Maverick, Gemma 4 31B, and more Why NVIDIA NIM is the most underrated free resource: - One account unlocks 100+ models, no need to register everywhere - 40 RPM is generous enough for daily dev and testing - Direct China access, no proxy needed - OpenAI-compatible API, base_url is https://integrate.api.nvidia.com/v1 - NVIDIA aggregates top models as infrastructure, not a model maker - New models added fast, often the first platform to offer free trials Best for: model comparison, prototyping, Agent tool calls, multi-model routing.
Ollama is recorded as China-friendly: Runs locally, no network needed. May need proxy to download models (or use Chinese mirrors).. Useful when you need access without complex network setup.
Ollama API has a recorded free trial: Unlimited; rate limit: Local.
Ollama has a recorded free tier: Unlimited (runs locally). Good for testing before upgrading.
Ollama is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
Ollama is recorded as supporting OpenAI-compatible API access. Free/trial info: Unlimited. Useful for low-cost testing by swapping SDK base_url.
Get Google AI Pro subscription free for 1 year via OnePass platform, worth $240/year ($19.99/mo × 12 months). How to: - Register OnePass account - Bind a Visa virtual card (no actual charge) - Claim Google AI Pro subscription on OnePass - Valid for 1 year, cancel before expiry Google AI Pro includes: - Unlimited Gemini 2.5 Pro (normally $19.99/mo) - 2TB Google One storage - AI features across Google apps (Gmail, Docs, Sheets) - Gemini Advanced premium features The easiest way to get Google AI Pro for free. No cost at all. Perfect for users who want Gemini Pro without paying.
This open source tool is designed for AI agents to perform budget checks before API calls, preventing high bills from infinite loops or misconfigurations. It gained 560 downloads within 3 days of release, indicating strong developer demand for such protection. The tool is completely free and open source, suitable for any team using AI agents.
OpenAI 于2026年4月将GPT-4o免费层从每日10次提升至50次，无需绑定支付方式即可使用，支持文本和图像输入。
新注册用户可获得少量免费额度，用于体验其聚合的众多模型API（如 Claude、GPT、Llama 等）。额度有限，适合初步测试。
OpenRouter 聚合多模型 API，新注册用户赠送 $1 免费额度，可用于 GPT-4、Claude 3.5、Gemini 等模型，国内可访问，无需信用卡。
Paperspace has recorded free compute or trial credits: free GPU notebooks (6hr sessions). Useful for inference, deployment, or GPU experiments.
Pieces offers a free or open-source option: Free plan includes core features — snippet management, AI chat, local model support. Useful for low-cost developer testing.
Pydantic AI offers a free or open-source option: Open-source framework, completely free to use. Useful for low-cost developer testing.
Qwen has released a workstation upgrade for the 3.6 27B model, supporting up to 5 concurrent users. This update targets multi-user collaboration scenarios, such as team development or research. Users need to configure the workstation environment to enable concurrency. The upgrade enhances the model's practicality, especially for small teams.
Qwen (Alibaba) is recorded as China-friendly: Alibaba Cloud service, direct access in China, very fast. Top-tier Chinese understanding. Qwen3.6-Plus coding near Claude Sonnet level.. Useful when you need access without complex network setup.
Qwen (Alibaba) API has a recorded free trial: 7000 万 tokens（新用户一次性）; rate limit: 按模型不同.
Qwen (Alibaba) has a recorded free tier: No explicit limit. Good for testing before upgrading.
Alibaba's Qwen3.6-Plus is the strongest Chinese coding model. New Bailian users get 70M free tokens (one-time). Coding ability close to Claude Sonnet 4.6, priced at only ¥2/M tokens.
Qwen (Alibaba) is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
Qwen (Alibaba) is recorded as supporting OpenAI-compatible API access. Free/trial info: 7000 万 tokens（新用户一次性）. Useful for low-cost testing by swapping SDK base_url.
Red Note AI is a Xiaohongshu AI post generator focused on persona-based writing, viral post references, and multi-model generation. New users get 3 free first generations.
Replicate 平台新用户注册即送$10免费额度，可用于运行多种开源模型（如Llama 3、Stable Diffusion），无需绑定信用卡，国内可注册使用。
平台托管大量 AI 模型，新用户注册可获得少量免费 GPU 时间，用于运行各种开源模型。超出后需付费。
Replicate 为新用户提供 $5 免费额度，可运行多种 AI 模型（图像生成、文本、语音等），国内可注册但需绑定支付方式。
Repomix offers a free or open-source option: Fully open-source and free, install via npm. Useful for low-cost developer testing.
Roo Code offers a free or open-source option: Open-source and free, bring your own API key. Useful for low-cost developer testing.
RunPod has recorded free compute or trial credits: $1 credits. Useful for inference, deployment, or GPU experiments.
SenseNova Token Plan beta is a lead for free DeepSeek-V4-Flash API access. Developers in China can test it for low-cost document handling, summarization, and simple agent subtasks. Current details come from a public article and platform entry; quotas and limits need re-verification.
A free website audit for SEO and GEO readiness. It checks whether your pages are easy for Google, AI search, and users to understand before launch or after a redesign.
A low-cost telco AI token package lead worth testing for effective cost per million tokens. Current info comes from a 2026-05-16 CLS screenshot: ¥1 for about 250k quota points, mobile-bill payment, and multiple model access. Verify quota conversion, supported models, and rate limits first.
SiliconFlow is recorded as China-friendly: Chinese service, direct access in China. Top choice for Chinese developers to access open-source model APIs.. Useful when you need access without complex network setup.
SiliconFlow 为新注册用户提供 2000 万 token 免费额度，支持 Llama、Qwen、DeepSeek 等多个开源模型，兼容 OpenAI API 格式，国内可直连，注册即送。
SiliconFlow API has a recorded free trial: ¥14; rate limit: Varies.
SiliconFlow 提供长期免费API额度，每月200万Token调用量，另赠送15元体验金可用于更高性能模型。支持多种开源模型（如Qwen、Llama、ChatGLM等），国内直连，注册即用。
SiliconFlow offers 14 open-source model APIs completely free, including Qwen, DeepSeek, Llama. Direct China access, fast speed, OpenAI compatible. The most convenient free AI API for Chinese developers.
注册 SiliconFlow 平台即送 2000 万 token，支持 Llama、Qwen、DeepSeek 等多种开源模型，国内直连，提供 OpenAI 兼容 API。
SiliconFlow has a recorded free tier: Varies by model. Good for testing before upgrading.
SiliconFlow is recorded as supporting OpenAI-compatible API access. Free/trial info: ¥14. Useful for low-cost testing by swapping SDK base_url.
iFlytek Spark is recorded as China-friendly: By iFlytek, direct access in China, industry-leading voice capabilities.. Useful when you need access without complex network setup.
iFlytek Spark API has a recorded free trial: 200万 tokens; rate limit: 5 RPM.
iFlytek Spark has a recorded free tier: No explicit limit. Good for testing before upgrading.
iFlytek Spark is recorded as supporting OpenAI-compatible API access. Free/trial info: 200万 tokens. Useful for low-cost testing by swapping SDK base_url.
StepFun is recorded as China-friendly: StepFun, direct access in China. Step 3.5 Flash is extremely fast.. Useful when you need access without complex network setup.
StepFun API has a recorded free trial: ¥10; rate limit: 5 RPM.
阶跃星辰为新注册用户提供 100万 token 免费 API 额度，支持 Step-2 万亿参数大模型，国内直连，注册即用，无需复杂审核。
StepFun has a recorded free tier: No explicit limit. Good for testing before upgrading.
StepFun is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
StepFun is recorded as supporting OpenAI-compatible API access. Free/trial info: ¥10. Useful for low-cost testing by swapping SDK base_url.
阶跃星辰 Step-2 大模型为新注册用户提供 100 万 token 的免费 API 调用额度，支持多模态和文本生成，国内直连，适合快速体验和开发测试。
Superset offers a free or open-source option: Open source and free to self-host locally. Useful for low-cost developer testing.
Superset is an integrated development environment (IDE) designed for the agent era, incubated by YC P26. It provides a complete toolchain to help developers build, debug, and deploy AI agent applications. The project is fully free and open source, with anyone able to access the GitHub repository for source code and contributions. As a newly launched product on its first day, Superset aims to lower the barrier to agent development, enabling more developers to get started quickly.
Tabnine offers a free or open-source option: Free plan offers basic completions using local small model, code stays on your machine. Useful for low-cost developer testing.
TEN Framework offers a free or open-source option: Open-source framework, completely free to use. Useful for low-cost developer testing.
腾讯混元大模型为开发者提供每月 100 万 token 的免费 API 调用额度，支持文本生成、对话等能力，国内开发者可直接使用微信/QQ 登录，无需绑定信用卡。
textgen has recorded free compute or trial credits: 免费算力/额度. Useful for inference, deployment, or GPU experiments.
Tiangong AI is recorded as China-friendly: By Kunlun Tech, direct access in China.. Useful when you need access without complex network setup.
Tiangong AI API has a recorded free trial: Free tier; rate limit: Varies.
Tiangong AI has a recorded free tier: No explicit limit. Good for testing before upgrading.
Tiangong AI is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
Together AI has recorded free compute or trial credits: $5 credits. Useful for inference, deployment, or GPU experiments.
Together AI gives new users $5 free credits for 200+ open-source model APIs. Highlights: - $5 free credits, enough for tens of thousands of API calls - FLUX image generation completely free, doesn't consume credits (hidden perk!) - Supports Llama 3.3 70B/405B, Mixtral 8x22B, Qwen 2.5, DeepSeek V3/R1 - Serverless and Dedicated deployment modes - OpenAI-compatible format - Fast inference, JSON Mode, Function Calling support FLUX free image generation is the biggest highlight: - FLUX.1 Schnell (fast, 1-4 step generation) - FLUX.1 Dev (high quality) - Completely free, unlimited, doesn't consume $5 credits - Quality comparable to Midjourney, great for batch product images and marketing assets Perfect for developers needing quality open-source model APIs plus free image generation.
unity-mcp offers a free or open-source option: Open-source and free to use, no usage limits. Useful for low-cost developer testing.
v0 offers a free or open-source option: Free plan with 200 messages per month for generating and iterating UI components. Useful for low-cost developer testing.
Vast.ai has recorded free compute or trial credits: $1 credits. Useful for inference, deployment, or GPU experiments.
VideoCaptioner offers a free or open-source option: Fully open-source and free, no usage limits when running locally. Useful for low-cost developer testing.
Vidu is recorded as China-friendly: Chinese platform, direct access in China, fast speed.. Useful when you need access without complex network setup.
Vidu API has a recorded free trial: $1; rate limit: N/A.
Vidu has a recorded free tier: 80 credits/month. Good for testing before upgrading.
字节跳动火山引擎提供的豆包大模型 API，新用户通常有一定量的免费 tokens 额度，国内可直接使用且稳定。
Warpdrv is a newly released open-source Llama.cpp launcher designed for daily-driving Qwen 35b and 27b models on Strix Halo and RTX Pro hardware. The project is completely free, and users can obtain the code directly from Reddit or GitHub. It simplifies the local LLM deployment process, suitable for users with compatible hardware for local inference.
Windsurf offers a free or open-source option: Free plan with unlimited basic completions + limited Cascade premium requests per month. Useful for low-cost developer testing.
wx_channels_download packages local proxying, stream parsing, and WeChat WebView button injection into a usable open-source tool. Its real value is not just downloading videos, but turning WeChat Channels content into assets that can be archived, transcribed, analyzed, and repurposed.
01.AI (Yi) is recorded as China-friendly: By 01.AI, direct access in China. Yi-Lightning is very cost-effective.. Useful when you need access without complex network setup.
01.AI (Yi) API has a recorded free trial: ¥10; rate limit: 5 RPM.
01.AI (Yi) has a recorded free tier: No explicit limit. Good for testing before upgrading.
01.AI (Yi) is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
01.AI (Yi) is recorded as supporting OpenAI-compatible API access. Free/trial info: ¥10. Useful for low-cost testing by swapping SDK base_url.
ChatGLM (Zhipu AI) is recorded as China-friendly: Chinese platform, direct access. GLM-4-Flash is free.. Useful when you need access without complex network setup.
注册智谱AI开放平台即送 100 万 token，可用于 GLM-4 系列模型，支持文本和图像生成，国内开发者直接使用，无需翻墙。
ChatGLM (Zhipu AI) API has a recorded free trial: 500万 tokens; rate limit: 5 RPM.
智谱AI 为新注册用户提供 100万 token 的免费 API 额度，可用于 GLM-4、GLM-4V 等模型，国内直连，支持 Python 和 HTTP 调用。
智谱 AI 为新注册用户提供 500 万免费 tokens，支持 GLM-4 系列模型，国内直接使用，无需翻墙，注册即送。
ChatGLM (Zhipu AI) has a recorded free tier: No explicit limit. Good for testing before upgrading.
智谱AI为GLM-4系列模型提供注册即送18元免费API额度，支持对话、代码生成等，国内开发者可直接使用，无需海外工具。
智谱 AI 为新注册开发者提供 500 万 token 免费额度，可用于 GLM-4、GLM-4V 等最新模型，国内直接使用，支持手机号注册，无需海外支付方式。
智谱AI为新注册用户提供500万Token免费额度（含GLM-4、GLM-4V等多模态模型），额外赠送100元API体验金，可用于更高阶模型调用。国内手机号直接注册，无需海外支付方式。
Zhipu GLM is a strong free API option for China-based developers today: registration is local-friendly, access is stable, and the API can be used in an OpenAI-compatible style. It is useful for Chinese customer support, knowledge-base QA, content generation, and multimodal experiments.
智谱AI 为新注册用户提供 100 万 token 的免费调用额度，同时赠送 100 元体验金，可用于 GLM-4、GLM-4V 等模型，支持国内直连，适合开发者和学生使用。
智谱 AI 为新用户提供 100 万 token 免费额度，可用于 GLM-4 系列模型（含 API 和 Web 端），国内直接注册使用，无需海外支付方式，适合中文场景开发。
智谱 AI 为开发者提供 GLM-4、GLM-3-Turbo 等模型的免费 API 调用额度，每月 100 万 Token，注册即享，支持国内网络直接使用，适合个人开发者和中小企业测试集成。
智谱 AI 为注册用户提供免费 100 万 token 额度，可用于 GLM-4、GLM-4-Flash 等模型 API 调用，国内开发者可直接使用，支持 Python SDK 和 OpenAI 兼容接口。
ChatGLM (Zhipu AI) is recorded as open-source or providing open model resources. Useful for local deployment, customization, and low-cost evaluation.
ChatGLM (Zhipu AI) is recorded as supporting OpenAI-compatible API access. Free/trial info: 500万 tokens. Useful for low-cost testing by swapping SDK base_url.
智谱 AI 为新注册用户提供 500万 Token 免费额度，可用于 GLM-4、GLM-4V 等模型 API 调用，国内直接访问，支持微信/支付宝实名认证。
🎁 Free Resource Pack
Free API credits list, AI business case studies, payment stack, risk checklist, and a monetization roadmap.