AI-generated: This article was created and published automatically by LinkLoot and was not substantively reviewed by a human editor.

Cloudflare adds DeepSeek V4’s 1M-context models to Workers AI

Cloudflare added DeepSeek V4 Pro and DeepSeek V4 Flash to Workers AI on August 14, making the models available through Workers AI bindings, REST, OpenAI-compatible endpoints, and AI Gateway. Cloudflare says they are its first Workers AI models with a full 1,048,576-token context window.

The release matters because long-context agent work is moving from specialist model endpoints into mainstream developer platforms. For builders already using Workers, this adds a large-context reasoning option without wiring a separate DeepSeek account into every app.

Key takeaways

• Cloudflare now lists @cf/deepseek-ai/deepseek-v4-pro-0813 and @cf/deepseek-ai/deepseek-v4-flash-0731 on Workers AI.
• Both models support reasoning, function calling, and a 1,048,576-token context window according to Cloudflare’s changelog.
• Access requires a Workers Paid plan or prepaid AI Gateway credits.
• DeepSeek’s official model card says V4 Pro 0813 supersedes the preview model and improves agentic performance.
• Independent benchmark context from Artificial Analysis lists DeepSeek V4 Pro 0813 as an open-weight, text-only reasoning model with 1.0M context.

What Cloudflare changed

Cloudflare’s changelog positions the two DeepSeek models as production routes for long-horizon agent workflows, large codebases, and multi-step reasoning. The hosted model pages expose usage through the same Workers AI surfaces developers already use: env.AI.run(), REST calls, OpenAI-compatible chat completions, and AI Gateway routing.

The Pro route is framed as the higher-capability reasoning model. Flash is the faster, lower-cost sibling and replaces the earlier preview with improved agentic capability.

Why the 1M-token window matters

A million-token context window does not guarantee perfect recall, but it changes what developers can attempt in one session. Whole repositories, long compliance files, historical support logs, and multi-step agent traces can fit into a single request more often, reducing brittle chunking and retrieval work.

That convenience has a cost side. Long prompts can still be expensive, slow, or noisy, and large context does not replace evaluation. Teams should test recall, tool-call reliability, and latency on their own workload before moving production agents to the new routes.

DeepSeek V4 Pro has moved past preview

DeepSeek’s Hugging Face model card describes DeepSeek-V4-Pro-0813 as the official release that supersedes the preview version. The card lists benchmark improvements for agent and coding tasks, including Terminal Bench, NL2Repo, Cybergym, DeepSWE, Toolathlon-Verified, and AutomationBench.

Artificial Analysis separately lists the model as released on August 13, 2026, with 1.6 trillion total parameters, 49 billion active parameters, a 1.0M-token context window, and open weights. Those details make the Cloudflare launch more than a routine catalog addition: it is a fast hosted path to a newly updated model family.

Access and migration notes

Cloudflare says both models require either a Workers Paid plan or prepaid AI Gateway credits. Existing Workers AI users should check plan eligibility, route naming, request tracing, and model-specific pricing before replacing another model in an agent workflow.

For teams comparing providers, the important distinction is that Cloudflare is not the only place the model exists. DeepSeek’s weights are available through Hugging Face, while Cloudflare packages access into its developer platform and AI Gateway control plane.

Source check

• Cloudflare’s changelog confirms the August 14 Workers AI availability, model IDs, 1,048,576-token context window, access requirements, and supported API surfaces.
• Cloudflare’s DeepSeek V4 Pro model page confirms the hosted model route and example usage.
• DeepSeek’s Hugging Face model card describes V4 Pro 0813 as the official release replacing the preview and lists benchmark context.
• Artificial Analysis independently tracks release date, context length, open-weight status, pricing, and benchmark positioning.
• DeepSeek’s arXiv technical report provides background on the V4 architecture and million-token efficiency claims.