Direct answer
Do not choose a production model from a static “latest models” ranking. Filter current models, groups, endpoints, and billing modes in the model and pricing catalog, then compare task success, latency, reliability, and actual charges on representative samples. This page uses the model snapshot regenerated on September 2, 2026. The current pricing source returned 249 models across 13 vendors. Re-check the candidates whenever the catalog changes.Current candidate entry points
These examples only mean that the ID is present in the current catalog. They do not prove access or compatibility for every account, parameter, or tool.
VIP per-call routes use
gpt-image-2.5-flare-vip or gpt-image-2.5-sunburst-vip at $0.03/call each, with the same integration as gpt-image-2-vip; see GPT Image 2.5.
A candidate is not a recommendation
The same model may:- be limited to a particular API key group;
- use different compatibility layers across endpoints;
- be billed by token, request, image, video, or another unit;
- be affected by upstream capacity, region, safety, or concurrency limits;
- support basic text but not tools, streaming, or multimodal fields.
Shared evaluation table
Record for every candidate:Model lifecycle
- A provider release does not mean LaoZhang API has integrated it.
- A catalog entry does not mean the current account has permission.
- Preview, stable, alias, and pinned IDs are defined by each provider.
- Legacy models can remain for migrations but should not be labeled as current defaults.
- Production migration should use parallel evaluation, gradual rollout, and rollback.