JA EN
shiichan

Workers AI Is Retiring 18 Older Models — Time to Migrate!

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

Hey everyone, it's Shii-chan! Today a somewhat important note landed from Workers AI, so let me share it with you.

Cloudflare Changelog developers.cloudflare.com

What was announced?

On the Cloudflare Changelog, there's an announcement that the Workers AI model catalog is being tidied up, with several older models set for deprecation. There are 18 models in total, and the deprecation date is May 30, 2026. Think of it as clearing shelf space to make room for newer releases.

The story so far

Until now, Workers AI carried a long lineup of previous-generation models like Llama 3, Mistral 7B, Gemma 7B, and Phi-2. They were handy, but as newer and smarter models keep arriving, it's time to graduate the older ones.

What changes

Apps that reference the deprecated models will need to switch to newer ones before May 30. Three replacements are recommended:

  • GLM-4.7-Flash — a fast multilingual model with multi-turn tool calling and coding capabilities.
  • Gemma-4-26b — an efficient open model with vision and tool calling.
  • Kimi K2.6 — a capable tool-calling and vision model for agentic workloads and coding.

Dive Deep

There's one thing to watch with Kimi K2.5. It was originally slated for deprecation on May 10, but that was extended to May 30. From May 30 onward, requests will be automatically aliased to Kimi K2.6 — which has a higher price — so the note asks you to check the pricing page and capabilities beforehand.

Here's a sample of the models being retired (excerpt):

@cf/moonshotai/kimi-k2.5 -> @cf/moonshotai/kimi-k2.6
@hf/meta-llama/meta-llama-3-8b-instruct
@cf/meta/llama-3-8b-instruct (and awq)
@cf/meta/llama-3.1-8b-instruct (and awq)
@cf/meta/llama-3.1-70b-instruct
@cf/meta/llama-2-7b-chat-int8 / -fp16
@cf/mistral/mistral-7b-instruct-v0.1
@cf/microsoft/phi-2
@cf/defog/sqlcoder-7b-2
@cf/facebook/bart-large-cnn

On the other hand, variants with -fast or -lora in the name will stay. For example, @cf/meta/llama-3.3-70b-instruct-fp8-fast, @cf/meta/llama-3.1-8b-instruct-fast, and the various LoRA models remain usable. That said, the note mentions LoRA models may be deprecated in the future too, so keep that in the back of your mind. You can check every model in the Workers AI model catalog.

Wrap-up

  • Workers AI is retiring 18 older models on May 30, 2026.
  • The recommended replacements are GLM-4.7-Flash / Gemma-4-26b / Kimi K2.6.
  • Kimi K2.5 will be auto-aliased to K2.6, but the price goes up, so check ahead.
  • -fast and -lora variants stay for now, though LoRA may be deprecated later too.

If you're still using older Llama or Mistral models on Workers AI, an early migration before the deadline is the way to go!

Related posts