Repository navigation
Agent Network: add gpt-6-luna to the OpenAI catalog before gpt-4o shuts down (Jan 14, 2027) #8664
prasadrandika
started this conversation in
Ideas & Feature Requests
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Before posting
Product area
Management service / API, Dashboard / Admin UI
Problem or use case
OpenAI shuts down
gpt-4o,gpt-4o-mini,gpt-4.1andgpt-4.1-miniin the API on January 14, 2027, and listsgpt-6-lunaas the recommended replacement (deprecations page).The Agent Network OpenAI catalog (
catalog.go:319-338) prices the retiring models but has no GPT-6 or GPT-5.6 entries. Requests togpt-6-lunastill route (on a catch-all provider, or once the model is added), but the model shows as unpriced. So after January 14, users moving fromgpt-4otogpt-6-lunawill see its cost tracked as $0 in Usage & Logs, unless an operator adds the rates by hand.Proposed solution
Add
gpt-6-lunato the OpenAI catalog with OpenAI's published Standard rates (gpt-6-luna model page):Context window: 1,050,000 tokens.
{ID: "gpt-6-luna", Label: "GPT-6 Luna", InputPer1k: 0.0001, OutputPer1k: 0.0005, CachedInputPer1k: 0.00001, ContextWindow: 1050000},Alternatives or workarounds considered
Operators can set per-model prices on the provider, or seed them with
server.agentNetwork.pricingDefaultsFile. That works, but every operator has to notice the $0 figures and do it themselves.Community impact and priority
gpt-6-lunarequestImplementation ideas
One catalog entry, plus a test that
LookupDefaultpricesgpt-6-luna. Two questions for you:gpt-6-lunacache writes ($0.125 per 1M, 1.25x input), while OpenAI-shape providers carry only a cached-input rate. Should cache writes be left out for now?Are you willing to help?
Yes, I can submit a PR if the approach is accepted.
Additional context
Prices checked on November 19, 2026 against the model page linked above. The Flex table on OpenAI's pricing page lists
gpt-6-lunaat exactly 50% of these rates, which matches its note that Flex is half of Standard.🤖 Generated with Claude Code
All reactions