Review · AI tools
Cloudflare Workers AI review
Serverless inference for open-weight models on Cloudflare's network, called from Workers or a REST API and billed in Neurons.
By the TopToolAdvisor research desk · Updated November 21, 2026
- Features (35%)
- 7.9
- Ease of use (35%)
- 8.3
- Value (30%)
- 9.1
Workers AI is a niche pick that the other LLM API lists usually skip, and it fits one kind of team well: a small company whose site or app already runs on Cloudflare Workers. The model call sits in the same code and deployment as the app, through an AI binding, so there is no extra vendor, key store or network hop to manage. Billing is in Neurons at $0.011 per 1,000 above a free allocation of 10,000 Neurons a day, and the pricing docs, updated November 18, 2026, now show the equivalent token price for each model. The cheapest LLM is IBM Granite 4.0 H Micro at $0.017 input and $0.112 output per 1M tokens, GPT OSS 20B is $0.20 and $0.30, and GPT OSS 120B is $0.35 and $0.75. Newer large models such as GLM-5.3, Kimi K2.6 and DeepSeek V4 Pro need a paid billing method. There are no proprietary frontier models, so Workers AI is best for classification, extraction and short replies.
Who it is for
Small teams already on Cloudflare Workers who want inference next to their app code
Price
Free 10,000 Neurons per day; paid usage at $0.011 per 1,000 Neurons, from $0.017 per 1M input tokens.
@cf/ibm-granite/granite-4.0-h-micro $0.017 per M input tokens $0.112 per M output tokens
Quoted from the vendor pricing page on November 21, 2026.
Key features
- Free allocation of 10,000 Neurons per day on both Workers Free and Workers Paid
- Per-model token prices published next to Neuron prices
- Open models including GPT OSS 120B and 20B, Llama, Qwen, GLM, Kimi and DeepSeek V4
- Called through the AI binding in a Worker or the REST API
- Can be paid with prepaid AI Gateway credits
Strengths and weaknesses
- Strength: IBM Granite 4.0 H Micro lists at $0.017 per 1M input tokens, the lowest token price on this page
- Strength: 10,000 free Neurons a day, reset at 00:00 UTC, cover small internal tools
- Strength: No separate vendor account for teams whose app already runs on Workers
- Weakness: Usage above the free allocation needs the Workers Paid plan
- Weakness: Some models, including GLM-5.3, Kimi K2.6 and DeepSeek V4 Pro, need a paid billing method
- Weakness: Neuron-based billing takes a minute to translate into per-request costs, and limits stop requests with an error once exceeded
Key facts
- Free allocation: 10,000 Neurons per day
- Paid rate: $0.011 per 1,000 Neurons above the free allocation
- Example prices: GPT OSS 20B, $0.20 input and $0.30 output per 1M tokens
- Setup effort: Low on Workers; needs a Cloudflare account otherwise
Where it ranks
Cloudflare Workers AI is No. 7 in our Best LLM APIs verdict.
Alternatives we scored
Visit cloudflare.com · Questions or corrections: editorial@toptooladvisor.com