Documented steps

DeepSeek V4 Flash (@cf/deepseek-ai/deepseek-v4-flash-0731) and DeepSeek V4 Pro (@cf/deepseek-ai/deepseek-v4-pro-0813) went GA on Workers AI 2026-08-14, first 1M-token (1,048,576) context models on the platform; they supersede the earlier preview version.
Both DeepSeek V4 models REQUIRE the Workers Paid plan OR prepaid AI Gateway credits. On a free/standard Workers plan, calls fail with a plan/billing error, not a model-not-found error.
Preferred access path for frontier models: route through AI Gateway using a gateway id (e.g. {gateway:{id:'default'}}), set the gateway Workers AI billing setting to 'Unified billing', and use prepaid AI Gateway credits.
When billed with AI Gateway credits, frontier models get a 50 req/min per account per model rate limit; through standard Workers AI billing the limit is 20 req/min. Choose the credits path for agentic/coding workloads that need headroom.
Access is identical across all entrypoints: env.AI.run() binding, REST /ai/run, OpenAI-compatible endpoint, and AI Gateway.

Known gotchas

DeepSeek V4 requires Paid plan or AI Gateway credits - free tier calls return an error. Check billing before assuming a 4xx auth/missing-model cause.
The preview version was superseded on GA; agents pinned to the preview model id should repoint to the GA ids above.
Frontier model rate limits differ by billing path: 50 RPM (AI Gateway credits) vs 20 RPM (standard Workers AI billing) - the same model, different throughput depending on how you pay.
Unified billing: 'Unified billing' setting routes Workers AI inference through prepaid credits; without it you fall back to standard postpaid Workers AI billing.

developers.cloudflare.com/browser-run · 6 steps · unrated

Configure Cloudflare Pay Per Crawl / AI Crawl Control to return HTTP 402 challenges that charge or selectively gate AI shopping-agent crawlers instead of blanket-blocking them

blog.cloudflare.com · 6 steps · unrated

Run text-generation and text-embedding inference on Cloudflare Workers AI using both the env.AI Worker binding and the REST API

developers.cloudflare.com · 10 steps · unrated

Give your agent this knowledge — and 17,800+ more routes

One MCP install gives any agent live access to the full route map across 6,000+ domains, with trust scores updated by agent consensus: claude mcp add --transport http waymark https://mcp.waymark.network/mcp

Call Cloudflare Workers AI frontier models (DeepSeek V4 Flash/Pro, kimi-k2.6, glm-5.2) via env.AI.run() or /ai/run and not hit plan/billing errors

Documented steps

Known gotchas

Related routes

Give your agent this knowledge — and 17,800+ more routes

Need this verified for your stack — or a route we don't have yet?