---
title: Cloudflare Adds DeepSeek V4 With a 1M-Token Context
description: Workers AI's first million-token context window arrives via DeepSeek V4 Pro and Flash, with thinking mode and function calling.
date: 2026-08-15T00:00:00.000Z
category: software-news
tags: cloudflare, deepseek, workers-ai
---

Cloudflare added DeepSeek V4 Pro and DeepSeek V4 Flash to its Workers AI platform on August 15,
giving developers a 1,048,576-token context window, the first models on the platform to reach
that mark.

## What shipped

The two models, `@cf/deepseek-ai/deepseek-v4-pro-0813` and `@cf/deepseek-ai/deepseek-v4-flash-0731`,
are available now through Workers AI, according to
[Cloudflare's own Workers AI model documentation](https://developers.cloudflare.com/workers-ai/models/).
Both support reasoning (thinking mode), function calling, and vision, positioning them for
long-horizon agent tasks rather than single-turn chat. DeepSeek V4 Flash is the faster,
lower-cost option of the pair; V4 Pro is the larger model.

## Why the context window matters here

A million-token context window is large enough to hold an entire mid-sized codebase, a lengthy
legal contract, or hours of transcribed audio in a single request, without the chunking and
retrieval workarounds developers currently use to fit long documents into smaller context
windows. Cloudflare frames the addition around long-horizon agent workflows and large-codebase
tasks specifically, not general chat.

## Why it matters

This is a platform-integration story more than a model story: DeepSeek already shipped V4 Pro and
Flash as its own release, and what's new here is Cloudflare making them available directly inside
Workers AI alongside its existing model catalog and billing. That distinction matters for
developers already building on Cloudflare's edge network, since it likely means one fewer
external API integration and one fewer vendor relationship to manage for teams that want a
long-context model without leaving the platform they're already deployed on. Whether developers
actually need a million tokens of context for most real workloads, versus a well-designed
retrieval pipeline over a smaller window, remains the open question a raw context-length number
doesn't answer on its own.

What to watch: real-world latency and cost at that context length once developers start using it
in production, not just the headline token count.

More [Tech Software](/software-news/) coverage, or everything tagged [cloudflare](/tag/cloudflare/).