AI DEAL COLLECTION
Free or trial APIs that can be called through OpenAI-compatible SDKs or base URLs for low-cost testing.
Free or trial APIs that can be called through OpenAI-compatible SDKs or base URLs for low-cost testing. 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.
DeepSeek announced R1 model API pricing at $0.14/M tokens input and $0.28/M tokens output, highly competitive.
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。
Groq increased free tier rate limit to 60 requests per minute, suitable for dev testing.
Hugging Face launched a free inference API supporting multiple open-source models, no credit card required, with 30,000 free inference requests per month.
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。
OpenAI announces a significant price cut for GPT-4.1 API, with input price reduced to $2/M tokens and output to $8/M tokens, offering better value than GPT-4o for large-scale API usage.
OpenAI announced a significant price drop for GPT-4.1 API, with input price reduced to $2/1M tokens and output to $8/1M tokens, offering better value than GPT-4o.
OpenAI announces a significant price reduction for GPT-4o mini API, with input price dropping to $0.15/M tokens and output to $0.60/M tokens, offering developers a more cost-effective AI service.
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。
SiliconFlow gives 20 million free tokens to new users, supporting multiple models.
SiliconCloud by SiliconFlow offers a 14-day free trial for new users, granting 20 million tokens for all models on the platform.
SiliconCloud offers new users a 14 RMB coupon for API calls, valid for 30 days.
Anthropic for Startups is a high-confidence official path for startup API credits and priority rate limits, but it is not an unconditional signup bonus. It targets VC-backed startups working with Anthropic VC partners; the credit amount is not publicly fixed and depends on Anthropic approval.
Anyscale API has a recorded free trial: $10 free credits; rate limit: 30 RPM.
Anyscale is recorded as supporting OpenAI-compatible API access. Free/trial info: $10 free credits. Useful for low-cost testing by swapping SDK base_url.
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。
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 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。
Cerebras API has a recorded free trial: 1M tokens/day; rate limit: 30 RPM / 60K TPM / 1M TPD.
Cerebras uses proprietary WSE chips for the world's fastest inference (2000+ tokens/s, 20x faster than GPU). Free tier: 1M tokens/day, 30 RPM, no credit card. Models: Llama 3.3 70B, Llama 3.1 8B, Qwen 3.5, and more. OpenAI-compatible API. Best for latency-sensitive use cases: real-time chat, streaming, Agent tool calls. Competes with Groq on speed, but with a larger daily token budget.
Cerebras is recorded as supporting OpenAI-compatible API access. Free/trial info: 1M tokens/day. Useful for low-cost testing by swapping SDK base_url.
ChatGPT (OpenAI) is recorded as supporting OpenAI-compatible API access. Free/trial info: $0. Useful for low-cost testing by swapping SDK base_url.
Anthropic API API has a recorded free trial: $5; rate limit: 5 RPM.
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 $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.
Cohere API has a recorded free trial: 1000 calls/month; rate limit: Trial rate limits.
新用户注册 Cohere 平台即获 $10 免费 API 额度，可用于 Command R+、Embed 等模型，支持 RAG 和分类任务，国内需科学上网。
Cohere offers a free Trial API Key with 1,000 calls/month across all models: - Command R+: top RAG and chat model - Rerank: document reranking for RAG pipelines - Embed: multilingual text embeddings No credit card required, resets monthly. Great for prototyping RAG projects. Note: Trial Key is not permitted for production use.
Cohere 提供每月 100 万 token 免费额度，支持 Command R+、Embed 等模型，API 稳定，国内需科学上网，适合 RAG 和文本生成场景。
Cohere 近期将免费试用额度从 40 万 token 提升至每月 100 万 token，支持 Command R、Embed 等模型 API，注册即享，国内需科学上网访问。
Coze (ByteDance) API has a recorded free trial: Free tier; rate limit: Varies.
Databricks announces the integration of OpenAI's GPT-5.5 model into its enterprise agent workflows. The model is designed for complex tasks, supporting multi-step reasoning and automated actions. Enterprise users can directly invoke it through the Databricks platform without additional configuration. This update marks a further expansion of OpenAI models in enterprise applications.
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。
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 increased free user daily conversation limit from 50 to 100, offering more free usage quota.
DeepSeek continues to offer free API credits, new users receive 5 million tokens upon registration, allowing immediate use without payment.
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。
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 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.
Doubao (ByteDance) API has a recorded free trial: 50万 tokens; rate limit: 5 RPM.
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.
ElevenLabs API has a recorded free trial: 10K chars/month; rate limit: Varies.
ERNIE Bot (Baidu) API has a recorded free trial: Free tier; rate limit: 5 RPM.
fal.ai API has a recorded free trial: Promotional credits; rate limit: N/A.
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。
Fireworks AI API has a recorded free trial: $1 free credits; rate limit: 600 RPM.
Fireworks AI is recorded as supporting OpenAI-compatible API access. Free/trial info: $1 free credits. Useful for low-cost testing by swapping SDK base_url.
FLUX (Black Forest Labs) API has a recorded free trial: Free via platforms; rate limit: Varies.
The Gemini API free tier is suitable for developers, small projects, and prototypes. Actual free rate limits vary by model, project, and billing tier, so users should confirm current limits in AI Studio.
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 AI (Gemini) API has a recorded free trial: 免费 API 无需信用卡; rate limit: 15 RPM (Flash).
Google 最新 Gemini 2.5 Pro 模型提供免费 API 层，每分钟最多2次请求，无需付费即可体验长上下文推理能力，适合开发测试和小型应用。
Google has updated the Gemini free tier quota. The Gemini 2.5 Flash model is now free on AI Studio with a rate limit of 30 requests per minute.
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，支持 Gemini 1.5 Pro 和 Flash 模型，每分钟最多 60 次请求，无需付费即可使用多模态能力，国内需代理访问。
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。
Google increased Gemini free tier context from 32k to 1M tokens and raised daily request limits, significantly enhancing the free user experience.
Google has announced the shutdown of its free search index, meaning AI applications and developers relying on web search can no longer access real-time search results for free. Traffic defense services like Cloudflare are also intensifying blocking of AI crawlers, further complicating web search. Users need to seek alternatives such as Bing API, DuckDuckGo, or self-built crawlers, though costs and technical barriers may increase.
On May 11, 2026, OpenAI released GPT-5.5 and the cybersecurity-focused GPT-5.5-Cyber model. This model series enhances trusted access capabilities, suitable for security analysis, threat detection, and automated response scenarios. The new models offer improved reasoning accuracy and safety, providing enterprises and security teams with a more reliable AI assistant.
OpenAI has released the GPT-5.5 Instant model, the latest iteration of the GPT series. This model is optimized for low-latency responses, suitable for applications requiring real-time interaction. Users can access it directly via the OpenAI API without additional application. Specific pricing and free tier details have not been announced yet; please follow official documentation for updates.
OpenAI has released the GPT-5.5 system card, marking the arrival of a new generation model. The model features significant improvements in reasoning, coding, and multimodal capabilities. Specific pricing and free tier details have not been announced yet, but it is expected to follow the tiered pricing strategy of the GPT series. Users can experience the new model via OpenAI API or ChatGPT.
OpenAI officially releases GPT-5.5 and GPT-5.5-Cyber models, the latest upgrade in the GPT series. GPT-5.5-Cyber is specifically designed for cybersecurity, offering enhanced trusted access control features for threat detection, vulnerability analysis, and more. The model helps enterprises better protect sensitive data and systems through strengthened security mechanisms.
Grok (xAI) API has a recorded free trial: $25/月; rate limit: Varies.
xAI's Grok gives $25 API credits monthly, auto-reset. Supports Grok-2 models with OpenAI compatible format. One of the highest monthly free API credits available.
Grok (xAI) is recorded as supporting OpenAI-compatible API access. Free/trial info: $25/月. Useful for low-cost testing by swapping SDK base_url.
Groq 提供基于 LPU 推理引擎的免费 API，支持 Llama 3、Mixtral 等模型，每日 1440 次请求限制，速度极快。需海外邮箱注册，国内可访问但需翻墙。
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 API has a recorded free trial: Free tier（永久免费）; rate limit: 30 RPM / 6000 TPM.
Groq is one of today's most useful free inference deals: the free tier lets developers test Llama, Mixtral, Gemma and other models through an OpenAI-compatible API. It is best for AI agents, RAG summarization, and low-latency chat prototypes. China access may require additional verification or a relay.
Groq 提供免费 API 额度，支持 Llama 3、Mixtral 等开源模型，推理速度极快，每日有限免费调用次数，注册即用，国内需科学上网。
Groq uses custom LPU (Language Processing Unit) chips for the fastest AI inference in the industry. Free models: - Llama 3.3 70B Versatile — 6000 TPM / 30 RPM - Llama 4 Scout 17B — 6000 TPM / 30 RPM - Llama 4 Maverick 17B — 6000 TPM / 30 RPM - Mixtral 8x7B — 5000 TPM / 30 RPM - Gemma 2 9B — 15000 TPM / 30 RPM - DeepSeek R1 Distill Llama 70B — 6000 TPM / 30 RPM Highlights: - 10x+ faster than GPU solutions, Llama 3.3 70B reaches 300+ tokens/sec - API keys start with gsk_, OpenAI-compatible - No total cap, rate-limited only - Requires proxy from China (use openllmapi.com)
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 increased free tier API rate limit from 30 to 60 requests per minute for more models.
Groq increased free tier API rate limits for more concurrent requests, ideal for developer testing and prototyping.
Groq deployed Meta's Llama 4 Scout and Llama 4 Maverick models with free API access.
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 Mixtral 8x7B 等模型的免费 API 访问，速率限制为每分钟30次请求，适合快速原型开发。国内需通过代理访问。
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。
Groq is recorded as supporting OpenAI-compatible API access. Free/trial info: Free tier（永久免费）. Useful for low-cost testing by swapping SDK base_url.
Hugging Face API has a recorded free trial: Free tier; rate limit: Varies.
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。
Tencent Hunyuan API has a recorded free trial: 100万 tokens; rate limit: 5 RPM.
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) 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) 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.
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) is recorded as supporting OpenAI-compatible API access. Free/trial info: $10 free credits. Useful for low-cost testing by swapping SDK base_url.
LM Studio API has a recorded free trial: Unlimited; rate limit: Local.
LM Studio is recorded as supporting OpenAI-compatible API access. Free/trial info: Unlimited. Useful for low-cost testing by swapping SDK base_url.
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 API has a recorded free trial: ¥15; rate limit: Varies.
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 API has a recorded free trial: Free tier; rate limit: 1 RPM.
Mistral AI 为新用户提供 500 万 token 免费 API 额度，支持 Mistral Large、Small 等模型，国内可注册但需海外邮箱。
Mistral AI’s official free API entry point is the Experiment plan: free for evaluation and prototyping, with limited rate limits; production or higher usage requires the Scale plan.
Mistral AI’s official free API entry point is the Experiment plan: free for evaluation and prototyping, with limited rate limits; production or higher usage requires the Scale plan.
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。
Mistral AI is recorded as supporting OpenAI-compatible API access. Free/trial info: Free tier. Useful for low-cost testing by swapping SDK base_url.
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。
Novita AI API has a recorded free trial: $0.50 free credits; rate limit: 60 RPM.
Novita AI is recorded as supporting OpenAI-compatible API access. Free/trial info: $0.50 free credits. Useful for low-cost testing by swapping SDK base_url.
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) 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.
OctoAI API has a recorded free trial: $10 free credits; rate limit: 60 RPM.
OctoAI is recorded as supporting OpenAI-compatible API access. Free/trial info: $10 free credits. Useful for low-cost testing by swapping SDK base_url.
Ollama API has a recorded free trial: Unlimited; rate limit: Local.
Ollama is recorded as supporting OpenAI-compatible API access. Free/trial info: Unlimited. Useful for low-cost testing by swapping SDK base_url.
OpenAI API has a recorded free trial: $5; rate limit: 3 RPM (free tier).
OpenAI announced a significant price reduction for GPT-4.1 API, with input price dropping to $2 per million tokens and output price to $8 per million tokens, approximately 26% cheaper than GPT-4o, greatly reducing developer costs.
OpenAI launches GPT-4.1 series API, approximately 26% cheaper than GPT-4o, with input at $2/M tokens and output at $8/M tokens. GPT-4.1 mini and nano are even more affordable for various use cases.
OpenAI announces significant price reduction for GPT-4.1 API, with input price dropping to $2 per million tokens and output to $8 per million tokens, approximately 50% lower than GPT-4o, greatly reducing developer costs.
OpenAI announced a significant price reduction for GPT-4.1 API, with input price dropping to $2 per million tokens and output price to $8 per million tokens, 26% cheaper than GPT-4o, offering developers more cost-effective AI capabilities.
OpenAI announces GPT-4.1 API price reduction, with input prices 26% lower and output prices 50% lower than GPT-4o; GPT-4.1 mini and nano are even cheaper.
OpenAI announces significant price reduction for GPT-4.1 API, with input price dropping to $2/M tokens and output to $8/M tokens, 26% cheaper than GPT-4o, greatly reducing developer costs.
OpenAI announces significant price reduction for GPT-4.1 API, with input price dropping to $2 per million tokens and output to $8 per million tokens, offering better value than GPT-4o for large-scale inference and generation tasks.
OpenAI announces price reduction for GPT-4.1 API series, with input price dropping to $2 per million tokens and output to $8 per million tokens, offering better value than GPT-4o.
OpenAI announced that the GPT-4.1 series models now support calling the code interpreter via API, allowing developers to leverage code execution for programming assistance, data processing, and analysis directly within their applications, significantly enhancing the model's utility in coding and data analysis scenarios.
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。
OpenAI 为 GPT-4o-mini 模型提供免费层，注册后每日可免费调用约100次，适合轻量级应用和测试。国内需通过代理访问。
OpenAI is recorded as supporting OpenAI-compatible API access. Free/trial info: $5. Useful for low-cost testing by swapping SDK base_url.
OpenAI has enhanced Structured Outputs for the GPT-4.1 series, improving JSON mode reliability and performance, enabling developers to obtain structured outputs more consistently.
OpenRouter API has a recorded free trial: Free models; rate limit: 20 RPM.
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 为新用户提供 $1 免费额度，同时提供多个永久免费模型（如 Mistral 7B、Llama 3 8B 等），支持统一 API 调用多种模型，国内需科学上网。
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。
OpenRouter 为新注册用户提供 $1 免费额度，可用于调用多种开源和商业模型（如 GPT-4、Claude、Llama 等），国内需代理访问。
OpenRouter is recorded as supporting OpenAI-compatible API access. Free/trial info: Free models. Useful for low-cost testing by swapping SDK base_url.
Perplexity AI is recorded as supporting OpenAI-compatible API access. Free/trial info: $0. Useful for low-cost testing by swapping SDK base_url.
Perplexity Pro 提供1个月免费试用，包含无限次搜索、高级模型（GPT-4、Claude 3等）和文件上传功能。需绑定支付方式，试用结束后自动续费（可取消）。国内可访问，但需科学上网。
Qwen (Alibaba) API has a recorded free trial: 7000 万 tokens（新用户一次性）; rate limit: 按模型不同.
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 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.
Replicate API has a recorded free trial: Free tier; rate limit: Varies.
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 提供每月 50 次免费推理额度，支持大量开源模型（如 Stable Diffusion、Llama、Whisper），国内需代理访问，适合模型测试和小型项目。
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。
SambaNova Cloud offers the world's only free LLaMA 3.1 405B API access. Core advantages: - LLaMA 3.1 405B (405 billion parameters) completely free — the largest free open-source model - The only platform globally offering free 405B access, bar none - Custom RDU (Reconfigurable Dataflow Unit) chip acceleration, ultra-fast inference - 30 RPM rate limit, no total cap — thousands of calls per day - API keys start with sn-, OpenAI-compatible format Supported models: - LLaMA 3.1 405B (flagship, best for complex reasoning) - Llama 3.3 70B (best value) - DeepSeek R1/V3 (671B MoE) - Qwen 2.5 72B - More models added regularly 405B vs 70B difference: - Significantly better complex reasoning (math, logic, multi-step) - Stronger long-text understanding (128K context) - Higher code generation quality - More precise instruction following Requires proxy from China (use openllmapi.com). Ideal for developers needing large model capabilities on a budget.
SambaNova API has a recorded free trial: Free tier（永久免费）; rate limit: 30 RPM.
SambaNova is recorded as supporting OpenAI-compatible API access. Free/trial info: Free tier（永久免费）. Useful for low-cost testing by swapping SDK base_url.
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 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 offers a 14-day free API trial for new users, supporting a variety of mainstream models, ideal for developers to quickly experience and test.
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。
SiliconCloud added multiple free models including DeepSeek-V3 and Qwen2.5 series, available for free API calls.
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 gives new users $10 API credits, valid for 30 days.
SiliconCloud offers 20 million free tokens for new users, supporting multiple models, ongoing promotion.
SiliconCloud gives new users 20 million free tokens for multi-model API calls, suitable for various AI application development.
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 API has a recorded free trial: 200万 tokens; rate limit: 5 RPM.
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 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 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。
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。
Tiangong AI API has a recorded free trial: Free tier; rate limit: Varies.
Together AI 为新用户提供 $25 免费 API 额度，可用于调用 Llama、Mixtral、Stable Diffusion 等开源模型，支持 OpenAI 兼容接口，国内需代理访问。
Together AI 为新用户提供每月 $25 免费额度，支持 Llama、Mistral、DeepSeek 等多种开源模型，国内需代理，适合模型微调和推理测试。
新注册用户获得 $25 免费 API 额度，支持 Llama 3、Mixtral、Falcon 等多种开源模型，兼容 OpenAI 格式，国内需代理访问，注册无需信用卡。
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.
Together AI offers $25 free API credits for new users, supporting 200+ open-source models. Key highlight: FLUX.1 Schnell Free image generation is completely free! - No credits consumed - Unlimited use - High-quality AI image generation - The only platform offering free high-quality AI image generation API LLM models: Llama 3.3 70B Turbo, Llama 4 Maverick, DeepSeek V3, Mixtral 8x22B, and 200+ more. API keys start with together-, OpenAI-compatible. base_url: https://api.together.xyz/v1 Requires proxy from China (use openllmapi.com).
Together AI API has a recorded free trial: $5（注册赠送）; rate limit: Varies by model.
Together AI is recorded as supporting OpenAI-compatible API access. Free/trial info: $5（注册赠送）. Useful for low-cost testing by swapping SDK base_url.
useknockout is an open-source project offering a free SOTA background removal and super-resolution API as an alternative to remove.bg and Topaz. It is MIT licensed and runs on the Modal platform, allowing users to utilize it within Modal's free tier. Suitable for developers and businesses needing image background removal or super-resolution processing.
Vidu API has a recorded free trial: $1; rate limit: N/A.
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。
01.AI (Yi) API has a recorded free trial: ¥10; rate limit: 5 RPM.
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.
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。
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 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.