AI models

Groq · Free

Rate-limited free access to GroqCloud; openai/gpt-oss-120b allows 30 requests and 8,000 tokens a minute, up to 1,000 requests and 200,000 tokens a day.

• 8Ktokens / minute
• 200Ktokens / day
• 30requests / minute

• Can't bill you
• No training on inputs

OpenRouter · Free

Free model variants (IDs ending in :free) are capped at 50 requests a day, raised to 1,000 after buying $10 of credits; free models whose providers train on prompts are only used once you switch on model training.

• 50requests / day

• Can't bill you
• Works in China
• No training on inputs

Cloudflare · Workers Free

10,000 Neurons a day at no charge on the Workers Free plan, shared across models; more usage needs Workers Paid, and some frontier models require a paid plan.

• 300requests / minute · LLM API
• 3Krequests / minute · Embeddings
• 10KNeurons / day

• Can't bill you
• No training on inputs

Mistral AI · Free mode

Free mode in Mistral Studio (formerly the Experiment plan) enables API access with no credit card; limits are shown in the console, and inputs may train models unless you opt out.

• Can't bill you
• No card
• Blocked in China

iFlytek Open Platform · Spark Lite

Spark Lite, iFlytek's lightweight language model (8K input, 4K output), is free to call through the Spark API on the iFlytek Open Platform. iFlytek's public pages give no token amount or rate limit for free use; requests over an app's authorized per-second or concurrency limit fail with flow-control errors.

• Commercial use
• Can't bill you
• No card
• Works in China

SiliconFlow · 免费模型

Selected open models are free to call after real-name verification, such as the Xing4.0-29B chat model and BGE embedding models; free models have fixed rate limits.

• Can't bill you
• Works in China
• No training on inputs

Google AI for Developers · Free

Free Gemini API access to selected models, including Gemini 2.5 Pro, Flash and Flash-Lite models, Gemma and embedding models; per-model rate limits are shown only in AI Studio, and free-tier content is used to improve Google products.

• Commercial use restricted
• Can't bill you
• Blocked in China
• Trains on inputs

NVIDIA · NVIDIA Developer Program

No-cost, rate-limited NIM endpoints on build.nvidia.com for NVIDIA Developer Program members; for prototyping and evaluation only, and inputs may be used to improve NVIDIA models.

• Non-commercial only
• Can't bill you
• Works in China
• Trains on inputs

Cohere · Trial

Free trial keys for Cohere models, limited to 1,000 API calls a month and 20 chat requests a minute; not permitted for production or commercial use.

• 1Krequests / month
• 20requests / minute
• 2KEmbed inputs / minute

• Non-commercial only
• Can't bill you
• Trains on inputs

Zhipu AI (BigModel) · 免费模型

GLM-4.7-Flash, GLM-4-Flash-250414 and other Flash models are free on the BigModel open platform; concurrency limits depend on the account tier and are shown in the console.

• Can't bill you
• Works in China
• Trains on inputs

Moonshot AI (Kimi) · 免费测试权益

New users who register with a mainland China phone number get a CNY 15 voucher after real-name verification, valid for 3 months and not usable on Kimi K3. Vouchers do not count toward cumulative top-ups, so voucher-only accounts stay on the lowest rate tier.

• Stops at the limit
• No card
• Works in China

Alibaba Cloud · 新人免费额度

New Bailian users get typically 1M free tokens per model in the Beijing region for 90 days, with no real-name check; unverified accounts stop when the quota runs out.

• 1Mtokens

• Stops at the limit
• No card
• Works in China
• No training on inputs

Volcengine · 免费推理额度

New Volcengine accounts get a one-off free quota of usually 500,000 tokens on each Ark model, for online inference only. Calls fail when a model's quota runs out; continuing needs real-name verification and activating the model, after which usage is billed per token unless Safe Experience Mode is on.

• 500Ktokens

• Commercial use restricted
• Stops at the limit
• No card
• Works in China
• No training on inputs

Cerebras · Free Trial

A one-time $5 credit, valid 30 days, after adding a payment method; gpt-oss-120b is limited to 5 requests and 30,000 uncached tokens a minute and 1M tokens a day.

• 30Ktokens / minute
• 1Mtokens / hour
• 1Mtokens / day

• Commercial use restricted
• Can't bill you
• Card required
• No training on inputs

Alibaba Cloud · Free quota for new users

New users get a free quota of typically 1M tokens per model in the Singapore region, valid 90 days; usage beyond it is billed unless Free Quota Only is turned on.

• 1Mtokens

• Works in China
• No training on inputs

Baidu AI Cloud · 新用户免费额度

New Qianfan users get 1M free tokens for each of 17 listed models, including ERNIE 4.5 Turbo, DeepSeek and Qwen3, valid 3 months and for online inference only.

• 1Mtokens

• Works in China
• No training on inputs

Tencent Cloud · 新人免费体验包

A one-time free trial on TokenHub, the Tencent Cloud model platform that replaced the old Hunyuan API: each main account can claim 1M free tokens per language model, including Tencent's own Hy (Hunyuan) models, valid for 1 year, while the campaign runs to December 31, 2026. Calls stop when a model's quota runs out unless pay-as-you-go billing is turned on.

• 1Mtokens

• Stops at the limit
• No card
• No training on inputs

Scaleway · Free Tier

New customers get a one-time 1,000,000 free tokens on the Serverless endpoints, shared by every token-billed chat, vision and embedding model, plus 60 minutes of audio transcription. A payment method is required and gives 200,000 tokens and 300 requests a minute per model. Usage beyond the free tokens is billed per token with no spending cap; prompts and outputs are not used for training.

• 1Mtokens
• 200Ktokens / minute
• 300requests / minute

• Commercial use
• Bills overage
• Card required
• No training on inputs