xAI retired eight Grok models from its API, a connection that lets developers plug a service into their own apps, on May 15, 2026, at noon Pacific Time. The company's migration guide lists grok-4-1-fast-reasoning, grok-4-1-fast-non-reasoning, grok-4-fast-reasoning, grok-4-fast-non-reasoning, grok-4-0709, grok-code-fast-1, grok-3, and grok-imagine-image-pro. Developers who left an old model name in a live app may not notice, because requests can still succeed while the bill and the model's behavior quietly change.
xAI's documents say the old names still work after the cutoff. Requests to the retired reasoning models, which work through a problem step by step before they answer, now go to grok-4.3 set to low reasoning effort. Requests to the non-reasoning models go to grok-4.3 with reasoning effort set to none. The coding model grok-code-fast-1 now goes to grok-build-0.1, and the image model grok-imagine-image-pro goes to grok-imagine-image-quality.
xAI says requests that stay on a retired text model after May 15 are billed at grok-4.3's rates, not the old model's: $1.25 for about 750,000 words of input and $2.50 for about 750,000 words of output. The live price list shows the same pair for grok-4.3 when a request is under about 150,000 words, and $2.50 and $5.00 when it is longer.
xAI recommended changing the model name in an app's code before May 15 instead of relying on the redirect, so developers control the reasoning effort and the cost. For work that calls for step-by-step answers, it points developers to grok-4.3 with the effort level set explicitly, and for coding to grok-build-0.1. The company also says grok-4.3 can take in about 750,000 words at once and has four effort levels: none, low, medium and high.
Some write-ups and accounts of the email xAI sent customers described the cutoff as a hard stop after which the old names would no longer work. That reading conflicts with the migration guide, which says the old names redirect. Unless xAI changes its documents, a quiet redirect with grok-4.3 billing is the behavior they describe. Developers who care about cost or quality can avoid surprises by naming the model and effort level directly.