Skip to content

Make it easier for scripts and AI agents to find the current list of supported AI Gateway models #32084

Description

@zeke

Existing documentation URL(s)

What changes are you suggesting?

Model availability on AI Gateway changes pretty often — new Anthropic, OpenAI, and Workers AI models show up regularly. It'd be great to have a more obvious, reliably-fetchable path for confirming "is model X supported right now, and what's the exact model string?" — something both humans and scripts/AI agents can lean on.

Right now there are a few different places with pieces of this info, and it's not always obvious how they relate to each other:

  1. The per-provider usage pages (e.g. anthropic/, openai/) show a model name in each code sample. It'd help to have a link from these pages to that provider's full model catalog, so it's clear there might be newer options beyond what's in the example.
  2. /ai/models/ is the full catalog, but it's a client-rendered, paginated page, which makes it harder to fetch or search programmatically (e.g. via curl).
  3. /ai/llms-full.txt is fetchable and does have current model info in it, but it's one big flat file without much internal structure, so finding a specific model means grepping and hoping you land on the right term.

A couple of ideas that might help:

  • Add a link from each provider's usage page to that provider's model catalog view (/ai/models/{provider}/).
  • A stable, crawlable, per-provider model index (JSON or markdown, in the spirit of llms.txt) that scripts/agents could fetch directly, rather than needing to scrape a JS-rendered page or grep a large flat file.

Additional information

For context, here's the scenario that prompted this: I was setting up Cloudflare AI Gateway with the Pi coding agent and asked an AI agent to check whether Claude Sonnet 5 and GPT-5.6 were supported. Based on what it found on the usage pages and a partial read of the model catalog, it concluded they weren't available yet. They actually already were — I had to give it the exact URLs myself (claude-sonnet-5, gpt-5.6-luna, gpt-5.6-terra, gpt-5.6-sol) before it confirmed they existed.

Once it had those URLs, it also confirmed the models were already present in llms-full.txt all along, by grepping it directly:

curl -s https://developers.cloudflare.com/ai/llms-full.txt \
  | grep -io "claude-sonnet-5\|gpt-5\.6[a-z-]*" | sort -u
claude-sonnet-5
GPT-5.6
gpt-5.6-luna
gpt-5.6-sol
gpt-5.6-terra

Sharing this mostly as a data point on where the discovery path could be smoother for automated tooling.

Metadata

Metadata

Labels

content:editRequest for content editsdocumentationDocumentation editsproduct:aiWorkers AI: https://developers.cloudflare.com/workers-ai/product:ai-gatewayAI Gateway: https://developers.cloudflare.com/ai-gateway/

Type

No type

Fields

No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions