CursorAI
CursorAI (listing points to Cloudflare Workers AI): serverless inference for 50+ open-source models on Cloudflare's global network.
Last verified:
What is CursorAI?
CursorAI, as listed here, points to Cloudflare Workers AI — a serverless AI inference platform that lets you run machine learning models on Cloudflare's global network without managing GPUs or infrastructure. It provides access to 50+ open-source models for tasks like text generation, image classification, object detection, embeddings, speech-to-text, and image generation, invokable directly from Workers or Pages code, or via the Cloudflare REST API using a single endpoint.
Key features include serverless pay-per-inference pricing with no idle costs, deployment across 200+ cities worldwide for low-latency inference, OpenAI-compatible API endpoints for text generation and embeddings, JSON Mode for structured output, integration with Cloudflare's ecosystem (AI Gateway, Vectorize, R2, D1), and support for multi-modal capabilities including LLMs, image generation, and audio processing. The platform handles provisioning, scaling, and latency optimization automatically.
Workers AI is designed for developers building AI applications at the edge — content platforms, social apps, voice agents, chat applications, search, and recommendation systems — and for teams wanting predictable pricing without GPU infrastructure management.
CursorAI pricing
Pricing model: Free
Workers AI is included in both Free and Paid Workers plans. Free tier: 10,000 Neurons per day at no charge, resetting daily at 00:00 UTC. Paid tier: $0.011 per 1,000 Neurons for any usage above the free allocation. Neurons measure GPU compute needed for your request across different models. LLM pricing varies by model size - for example, @cf/meta/llama-3.2-1b-instruct costs 2457 neurons per M input tokens and 18252 neurons per M output tokens. Image models price per step or tile (e.g., flux-1-schnell: 9.60 neurons per step). Audio models price per audio minute (e.g., whisper: 41.14 neurons per audio minute). You can monitor Neuron usage in the Cloudflare Workers AI dashboard.
CursorAI pros
- Serverless pay-per-inference pricing with no idle costs
- 50+ open-source models in curated catalog
- Deploy to 330+ cities instantly with low latency
- No GPUs to manage or capacity planning required
- OpenAI-compatible API works with any OpenAI SDK
- Single API call to run any model from your code
- Native integration with AI Gateway, Vectorize, R2, D1
- JSON Mode for structured JSON output responses
- Multi-modal support: LLMs, images, audio, embeddings
- Daily free tier of 10,000 Neurons resets at UTC midnight
- Automatic scaling and latency optimization handled automatically
- No cold starts or region complexity
- Predictable pricing without surprises
- Enterprise-grade reliability on Cloudflare infrastructure powering 20% of Internet
- Support for reasoning models like deepseek-r1 and function calling
CursorAI cons
- Limited to curated model selection, cannot deploy custom models on standard plans
- Custom model hosting requires enterprise plans with higher costs
- Potential cold start latency for infrequently used models
- Vendor lock-in to Cloudflare infrastructure ecosystem
- Pricing can be unpredictable for high-volume applications without monitoring
- Limited fine-tuning options compared to dedicated model hosting platforms
- Documentation and community support still developing vs established AI platforms
- Beta models may have lower rate limits while scaling
Frequently asked questions about CursorAI
What are Neurons and how do they work?
Neurons measure the GPU compute needed to perform your AI request across different models. They represent a standardized way to compare usage across model types like tokens for LLMs, audio seconds for speech models, images for image generation, and steps. Each model's input format maps to Neurons - for example, Llama 3.2 1b uses 2457 neurons per M input tokens. You get 10,000 Neurons free daily, then pay $0.011 per 1,000 Neurons above that.
How do I get started with Workers AI?
First, sign up for a Cloudflare account if you don't have one. Get your API token and Account ID by going to the Workers AI page in the Cloudflare dashboard, selecting Use REST API, creating a Workers AI API token with Read access, and copying your Account ID. Then you can run models via the REST API using the Execute AI model endpoint, or integrate directly from your Workers/Pages code using the env.AI binding.
What models are available in Workers AI?
Workers AI offers 50+ open-source models including Meta's Llama 3.1/3.2/3.3 variants (8b, 11b, 70b), Mistral 7b, DeepSeek R1 distill Qwen 32b, OpenAI's GPT-OSS 120b and 20b, Qwen 3 Coder, Llama 4 Scout 17b, Google Gemma models, FLUX image generation models, Whisper for speech-to-text, and embedding models like BGE small/base/large. Models cover text generation, embeddings, image generation, speech-to-text, classification, and object detection.
Can I use Workers AI with OpenAI SDK?
Yes, Workers AI supports OpenAI-compatible endpoints for text generation (/v1/chat/completions) and text embedding models (/v1/embeddings). You can use any OpenAI SDK to connect to Workers AI. The platform also supports OpenAI API schema through the /ai-gateway/chat-completion/ endpoint for accessing Workers AI models.
What is JSON Mode and which models support it?
JSON Mode forces Workers AI text generation models to return valid JSON output using response_format or JSON schemas, enabling structured output for applications. It's compatible with OpenAI's implementation. Supported models include: @cf/meta/llama-3.1-8b-instruct-fast, @cf/meta/llama-3.1-70b-instruct, @cf/meta/llama-3.3-70b-instruct-fp8-fast, @cf/meta/llama-3-8b-instruct, @cf/meta/llama-3.1-8b-instruct, @cf/meta/llama-3.2-11b-vision-instruct, @hf/nousresearch/hermes-2-pro-mistral-7b, @hf/thebloke/deepseek-coder-6.7b-instruct-awq, and @cf/deepseek-ai/deepseek-r1-distill-qwen-32b. Note: JSON Mode currently doesn't support streaming.
What are the rate limits for Workers AI?
Rate limits are per task type: Text Generation has 300 requests per minute (with some models like mistral-7b at 400 rpm, phi-2 at 720 rpm), Text Embeddings has 3000 rpm, Image Classification has 3000 rpm, Text-to-Image has 720 rpm, Speech-to-text has 720 rpm, and Object Detection has 3000 rpm. Beta models may have lower rate limits. Limits apply per task type and some models have specific per-model limits.
How does Workers AI integrate with other Cloudflare products?
Workers AI tightly integrates with Cloudflare's ecosystem: AI Gateway for analytics, caching, rate limiting, request retries, and model fallback; Vectorize for vector database enabling semantic search and recommendations; Workers for serverless application deployment; Pages for full-stack applications; R2 for egress-free storage of large datasets; D1 for serverless SQL databases; KV for key-value storage; and Durable Objects for globally distributed coordination. This unified platform reduces tool sprawl.
Can I stream responses from Workers AI models?
Workers AI supports streaming for text generation models. You can use the Vercel AI SDK for JavaScript and TypeScript codebases which enables streaming text generation, tool calls, and structured output. However, JSON Mode currently doesn't support streaming - when using JSON Format with response_format, streaming is not available.
What happens if I exceed my free tier limit?
If you exceed your 10,000 Neurons daily free allocation, further operations will fail with an error unless you upgrade to the Workers Paid plan. On Workers Paid, you're charged $0.011 per 1,000 Neurons for any usage above the free tier. All limits reset daily at 00:00 UTC, so your free tier refreshes every 24 hours. You can monitor your Neuron usage in the Cloudflare Workers AI dashboard.
How do I access Workers AI from my code?
You can access Workers AI from your Workers or Pages code using the env.AI binding. In your Worker, call await env.AI.run(modelName, { messages: [...] }) where modelName is the model identifier like '@cf/meta/llama-3.1-8b-instruct'. You can also use the REST API by authenticating with your API token and making requests to the Execute AI model endpoint. For Vercel AI SDK, install workers-ai-provider and add an AI binding in your Wrangler file.