All guidesLaunch checklists by stack and tool
On this page

Lovable change

LovableGemini

Lovable AI model retirement dates: Gemini 3.6 Flash on November 19, 2026, Gemini 3.7 Flash on January 28, 2027, and what your app gets after

By Ohad Krispin, founder of VibeRaven · LinkedIn

Lovable stops serving Gemini 3.6 Flash (google/gemini-3.6-flash) to your app's AI features on November 19, 2026 and Gemini 3.7 Flash (google/gemini-3.7-flash) on January 28, 2027, at 00:00 UTC. Lovable's deprecated models page says requests for both then go to Gemini 3.8 Flash, so the feature keeps working on a model you did not pick, which can cost a different amount and answer differently; the changelog entries that announced them say both models stop working on those dates. Some retired models are not replaced: after GPT-4o mini TTS retires on January 6, 2027, and GPT-4o Transcribe and GPT-4o mini Transcribe on February 26, 2027, requests fail with an error that names the model and its date. Lovable does not change the model in your code on its own: ask it to replace the deprecated models, then publish, because until you publish only the preview uses the new model. Checked against Lovable's docs and changelog on 2026-10-25.

Find the models your app uses

Open More, AI in the project: the AI activity list shows the model each AI request used. After a retirement date it shows the replacement for requests Lovable redirected, so look before the date. You can also ask Lovable in the project chat which models the app's AI features use. In a clone of the repository Lovable syncs to GitHub, search the code for the model ids. Lovable's changelog writes them as provider/model, such as google/gemini-3.6-flash; its deprecated models page gives display names only.

grep -rn --exclude-dir=node_modules -e "gemini-3.6-flash" -e "gemini-3.7-flash" -e "gemini-2.5" -e "gpt-4o-mini-tts" -e "gpt-4o-transcribe" -e "gpt-4o-mini-transcribe" .

Replaced or failing: what happens on the date

At 00:00 UTC on the retirement date Lovable stops sending requests to the model. For a replaced model, Lovable sends the request to the replacement and your app keeps working without a code change, at that model's price and with its answers. Lovable says the replacement is the nearest current model, not always the newest, and suggests you pick the model yourself. For a rejected model, such as the speech models above, requests fail with an error that names the model and its retirement date, and the feature stops until you switch.

Switch to a current model

Ask in the project chat, with Lovable's own prompt below, or name the model you want. Lovable updates the backend code that calls the model. Test the AI features in the preview, then publish: until you publish, only the preview uses the new model.

Replace any deprecated AI models in this app with current ones.

Why a model is deprecated on Lovable before the provider retires it

Lovable can deprecate or retire a model on its own, even when the provider still offers it, for example when an agreement with a provider ends or when it stops supporting an older generation. A deprecated model keeps working in the apps that use it until its retirement date, and Lovable no longer picks it for a new AI feature unless no current model covers that job. Lovable emails the project owners whose apps call the model as soon as it has a retirement date.

Checklist

Before November 19, 2026

  1. 01Open More, AI, AI activity and write down every model the app's requests used.
  2. 02Compare each with the table above. Mark every model with a date before your next release, and every model whose requests fail.
  3. 03Ask Lovable to replace the deprecated models, or name the model you want for each feature.
  4. 04Test each AI feature in the preview: its output, its errors and the credits it uses per request.
  5. 05Publish, then check AI activity again: requests from the published app should show the new model.

Retirement table

Every deprecated model on Lovable, with its date

Copied from Lovable's deprecated models page on 2026-10-25. Lovable's dates decide when your app's requests change, not the provider's, and Lovable can retire a model before its provider does. A date in the past means the model is already retired.

Deprecated models and what requests get after the date

Lovable deprecated AI models, read 2026-10-25
ModelTypeRetirement dateAfter the retirement date
Gemini 2.5 ProChatOctober 20, 2026Requests go to Gemini 3.8 Flash
Gemini 2.5 FlashChatOctober 20, 2026Requests go to Gemini 3.8 Flash
Gemini 2.5 Flash LiteChatOctober 20, 2026Requests go to Gemini 3.1 Flash Lite
Gemini 3.6 FlashChatNovember 19, 2026Requests go to Gemini 3.8 Flash
GPT Image 1 MiniImageDecember 1, 2026Requests go to GPT Image 2.5 Flare
GPT-5ChatDecember 11, 2026Requests go to GPT-5.6 Sol
GPT-5 miniChatDecember 11, 2026Requests go to GPT-5.6 Terra
GPT-5 nanoChatDecember 11, 2026Requests go to GPT-5.6 Luna
GPT-4o mini TTSText to speechJanuary 6, 2027Requests fail
Gemini 3.7 FlashChatJanuary 28, 2027Requests go to Gemini 3.8 Flash
GPT-4o mini TranscribeSpeech to textFebruary 26, 2027Requests fail
GPT-4o TranscribeSpeech to textFebruary 26, 2027Requests fail
Gemini 2.5 Flash ImageImageMarch 15, 2027Requests go to Gemini 3 Pro Image
GPT-5.4 NanoChatApril 1, 2027Requests go to GPT-6 Luna
GPT-5.4 ProChatNot announcedRequests go to GPT-5.5 Pro
GPT-5.2ChatNot announcedRequests go to GPT-5.6 Sol
Gemini 3.5 FlashChatNot announcedRequests go to Gemini 3.8 Flash
Gemini 3.1 Flash ImageImageNot announcedRequests fail
Gemini 2.5 Flash Lite Preview TTSText to speechNot announcedRequests go to Gemini 2.5 Flash TTS

Related

Sources

Official docs behind this checklist

FAQ

FAQ for builders before launch

Does this change the model Lovable uses to build my app?

No. These are the models your deployed app calls through Lovable's AI gateway, billed as AI gateway usage, separate from the Lovable agent that edits your project.

Can I keep using a deprecated model until its date?

Yes. If you ask for one, Lovable tells you it is deprecated and offers a current model; if you ask again, it uses the model you asked for.