JA EN
shiichan

AI Search just gained 6 new text generation models!

Cloudflare Changelog ★★★★ View Markdown
#cloudflare#ai

Hey everyone, it's Shii-chan!

Cloudflare Changelog developers.cloudflare.com

What was announced?

Cloudflare's Changelog announced new Workers AI text generation models available in AI Search — six of them, all at once.

Here's what's new:

  • @cf/deepseek-ai/deepseek-v4-flash-0731 (1,048,576-token context window)
  • @cf/deepseek-ai/deepseek-v4-pro-0813 (1,048,576-token context window)
  • @cf/openai/gpt-oss-120b (128,000-token context window)
  • @cf/openai/gpt-oss-20b (128,000-token context window)
  • @cf/qwen/qwen3.8-27b (262,144-token context window)
  • @cf/moonshotai/kimi-k2.7-code (262,144-token context window)

The two DeepSeek models stand out with context windows over a million tokens.

Why it matters

AI Search combines vector search with text generation, RAG-style. The more text generation models you can choose from, the easier it is to pick the right one for your use case. This update brings a broad lineup of open-weight models — DeepSeek, GPT-OSS, Qwen, and Kimi — all at once.

What changes

You can now select any of these new models when creating or updating an AI Search instance, either from the dashboard or the API. Since they all run on Workers AI, you don't need to bring your own API key from another provider — it's all handled within your Cloudflare account.

Dive Deep

Context window sizes vary quite a bit across these models. The two DeepSeek models offer a massive 1,048,576 tokens, which is great if you want to feed in entire long documents. The GPT-OSS models sit at a more standard 128,000 tokens, while Qwen and Kimi land in between at 262,144 tokens. kimi-k2.7-code in particular sounds tailored for coding-related tasks based on its name.

The announcement doesn't mention specific pricing, so it's worth checking the Workers AI pricing page before you dive in.

Wrap-up

  • AI Search gained six new Workers AI text generation models
  • The lineup spans two DeepSeek models, two GPT-OSS models, Qwen, and Kimi
  • Context windows range from 128,000 to 1,048,576 tokens depending on the model
  • They run on Workers AI, so no extra provider key is needed — pick them right from the dashboard or API
  • Great news if you're building RAG on AI Search and want more model choices!

Related posts