Resources // LLM Tokens
Cloudflare Workers AI
Offer 10,000 Neurons/day free
Suits for Non-VC-backedCreatorsResearchersOpen-source
Updated Jun 2026
The offer
- Free allocation of 10,000 Neurons per day on both the Workers Free and Workers Paid plans (resets daily at 00:00 UTC)
- Neuron = Cloudflare's GPU-compute unit; serverless, pay only for what you use. Beyond the free allocation: $0.011 per 1,000 Neurons (requires the Workers Paid plan)
- Broad catalog billed in Neurons: LLMs (Llama 3.x / 4, DeepSeek-R1-distill, Mistral, Qwen, Gemma, gpt-oss), embeddings (BGE), image (FLUX), audio (Whisper), rerankers, and translation
- OpenAI-compatible endpoints, Workers Bindings, REST API, AI Gateway, function calling, and LoRA fine-tunes, running on Cloudflare's global edge network
- Light/embedding/audio workloads stretch furthest: Whisper transcription is roughly $0.03 per audio-hour
Who qualifies
- Any Cloudflare account (free) gets the 10,000 Neurons/day allocation: no application, no credit card for the free tier
- To exceed the daily free allocation you need the Workers Paid plan ($5/month), then it is $0.011 per 1,000 Neurons
- Limits reset daily and are hard: exceeding them fails further operations with an error until reset
- This is the always-on free inference tier, separate from "Cloudflare for Startups" (which grants larger time-limited credits to accepted startups)
Community Insights
Developers like Workers AI for cheap, serverless, edge-deployed inference. It comes up repeatedly for low-cost Whisper transcription (around $0.03 per audio-hour) and as one of the standard hosts now carrying open models like DeepSeek. The most common knock is a comparatively limited model selection versus dedicated inference shops, plus a general sense that Cloudflare’s broader Workers pricing runs steep next to newer clouds. Treat the 10,000-Neurons/day free tier as an easy on-ramp for testing; production volume needs the paid tier.
Best Practices (from community tips)
- Lean on it for embeddings, reranking, Whisper, and small/fast LLMs where 10k Neurons/day stretches furthest; send heavy chat to a paid tier or another host.
- Pick the
-fp8-fast/ quantized model variants. They cost far fewer Neurons per token than full-precision versions. - Use the OpenAI-compatible endpoints so you can A/B against other providers and migrate easily.
- Put AI Gateway in front for caching, rate-limiting, and spend visibility; cache repeated prompts to conserve the daily allocation.
- Design around the 00:00 UTC daily reset. The limits are hard-fail, so monitor Neuron usage in the dashboard.
- For production volume, budget the $0.011/1k-Neuron overage explicitly and consider applying to “Cloudflare for Startups” separately for larger credits.