Cloudflare Workers AI serves open models on its global edge via an OpenAI-compatible endpoint
/accounts/{id}/ai/v1. Add your Cloudflare key in Settings, then load your account’s full live catalog below — every task type (text, image, audio, video, embeddings, classification…), searchable and filterable. The curated text models further down are what the router picks from.| Model | Maker | Type | Context | $ In | $ Out | Speed t/s | Capability | Best for |
|---|---|---|---|---|---|---|---|---|
GLM-5.3 @cf/zai-org/glm-5.3 | Z.ai (Zhipu) | chat | 1M | $1.40 | $4.40 | 110 | Z.ai's coding flagship at 1M context. Same base model as GLM-5.2 — Z.ai states every gain comes from post-training — with large jumps on coding and cyber benchmarks. Requires a paid Workers plan. | |
DeepSeek V4 Pro (0813) @cf/deepseek-ai/deepseek-v4-pro-0813 | DeepSeek-AI | chat | 1M | $1.32 | $3.96 | 110 | DeepSeek's 1.65T-parameter flagship at 1M context, in the dated 0813 checkpoint Cloudflare serves. Cached input is $0.044 — a 30x discount that makes repeated-prefix agent loops far cheaper than the headline suggests. | |
GLM-5.2 @cf/zai-org/glm-5.2 | Z.ai (Zhipu) | chat | 262k | $1.40 | $4.40 | 110 | The base GLM-5.3 was post-trained from, still served in its own right at the same price. | |
Kimi K2.7 Code @cf/moonshotai/kimi-k2.7-code | Moonshot AI | chat | 262k | $0.95 | $4.00 | 110 | Moonshot's code-specialized K2.7 with vision. Sits below the 2.78T K3 but is served here at a fraction of a frontier price. | |
Kimi K2.6 @cf/moonshotai/kimi-k2.6 | Moonshot AI | chat | 262k | $0.95 | $4.00 | 110 | General-purpose K2.6, and the auto-alias target for the now-deprecated K2.5 — note Cloudflare aliases it upward in price, so K2.5 traffic silently costs more. | |
DeepSeek V4 Flash (0731) @cf/deepseek-ai/deepseek-v4-flash-0731 | DeepSeek-AI | chat | 1.3M | $0.44 | $1.32 | 140 | The longest context window in this catalog at 1,310,720 tokens, for a fifth of Pro's price. | |
GPT-OSS 120B @cf/openai/gpt-oss-120b | OpenAI (open-weight) | chat | 131k | $0.03 | $0.17 | 110 | Flagship open-weight reasoning model on Cloudflare’s edge — strong agentic + reasoning, low global latency. | |
GLM-5.3 Flash @cf/zai-org/glm-5.3-flash | Z.ai (Zhipu) | chat | 1M | $0.15 | $0.50 | 160 | The cheapest 1M-context model in this catalog by a wide margin, and the architecturally interesting half of the 5.3 pair: a new base model with 3:1 KDA linear attention to sparse full attention, NoPE MLA and 4-stream hyper-connections. Vision-capable. | |
Nemotron 3 Super 120B-A12B @cf/nvidia/nemotron-3-120b-a12b | NVIDIA | chat | 256k | $0.50 | $1.50 | 95 | NVIDIA's 120B/12B-active Nemotron 3, served per-token on Cloudflare rather than on NIM credits — the one place in this catalog a Nemotron carries a real price. | |
Llama 3.3 70B (fp8 fast) @cf/meta/llama-3.3-70b-instruct-fp8-fast | Meta | chat | 128k | $0.29 | $2.25 | 90 | Largest general-purpose Llama on Workers AI — quality-first chat, RAG and tool use at the edge. | |
Qwen2.5 Coder 32B @cf/qwen/qwen2.5-coder-32b-instruct | Alibaba Qwen | chat | 33k | $0.66 | $1.00 | 80 | Best-in-class open coding model on Workers AI — code generation, completion and repair. | |
Qwen3.8 27B @cf/qwen/qwen3.8-27b | Alibaba Cloud (Qwen) | chat | 262k | $0.45 | $3.20 | 130 | The same Qwen3.8 27B Groq serves, at roughly half Groq's input price and twice the context — a genuine cross-provider arbitrage the router can exploit. | |
Llama 4 Scout 17B @cf/meta/llama-4-scout-17b-16e-instruct | Meta | chat | 131k | $0.27 | $0.85 | 120 | MoE model with native vision + long context — multimodal chat and document understanding on the edge. | |
Mistral Small 3.1 24B @cf/mistralai/mistral-small-3.1-24b-instruct | Mistral AI | chat | 128k | $0.35 | $0.56 | 110 | Efficient 24B with vision + function calling — a strong balanced default for production agents. | |
Gemma 4 26B-A4B @cf/google/gemma-4-26b-a4b-it | chat | 256k | $0.10 | $0.30 | 175 | The cheapest vision-plus-reasoning model in the catalog — a 26B/4B-active MoE Gemma at $0.10 input. | ||
GPT-OSS 20B @cf/openai/gpt-oss-20b | OpenAI (open-weight) | chat | 128k | $0.20 | $0.30 | 190 | The small GPT-OSS on the edge. Note it costs more per input token here than on Groq ($0.075) but less per output token ($0.30 vs $0.30 — level), so the cheaper provider depends on your output ratio. | |
GLM-4.7 Flash @cf/zai-org/glm-4.7-flash | Z.ai (Zhipu) | chat | 131k | $0.06 | $0.40 | 180 | Cheap small-context GLM for high-volume classification and routing work. | |
SEA-LION v4 27B @cf/aisingapore/gemma-sea-lion-v4-27b-it | AI Singapore | chat | 128k | $0.35 | $0.56 | 120 | Southeast-Asian language specialist on a Gemma base — the strongest multilingual option here for SEA languages specifically. | |
Llama 3.1 8B (fast) @cf/meta/llama-3.1-8b-instruct-fast | Meta | chat | 128k | $0.05 | $0.20 | 220 | Cheap, fast 8B for high-volume chat, classification and SDR-style tasks at the edge. | |
Granite 4.0 H Micro @cf/ibm-granite/granite-4.0-h-micro | IBM | chat | 131k | $0.02 | $0.11 | 200 | The cheapest function-calling model in the catalog by roughly 3x, and a Mamba-2 hybrid rather than a plain transformer. For high-volume tool-routing where per-call cost dominates. | |
Llama 3.2 3B @cf/meta/llama-3.2-3b-instruct | Meta | chat | 131k | $0.05 | $0.33 | 300 | Tiny multilingual model for edge dialogue, routing and cheap summarization. | |
Llama 3.2 1B @cf/meta/llama-3.2-1b-instruct | Meta | chat | 60k | $0.03 | $0.20 | 400 | Ultra-light 1B for the cheapest, fastest classification, extraction and guardrail checks. |