On 1 December 2026, Google starts answering API requests for gemini-2.5-pro with gemini-3.8-flash and gemini-2.5-flash with gemini-3.5-flash-lite. The calls keep going through, but the answers come from a different model.
What changes
Google lists these models on its deprecations page. The dates and replacements on this page come from Driftline's registry, which is checked against that page several times a day, so they stay current if Google changes them.
| Model | Replacement | Moving to it |
|---|---|---|
gemini-2.5-pro | gemini-3.8-flash | Parameter changes |
gemini-2.5-flash | gemini-3.5-flash-lite | Parameter changes |
More than a new model name
For these models, changing the name isn't the whole job:
- Gemini 3 changes the request: remove temperature, top_p, top_k, candidate_count and prefilled model turns, replace thinking_budget with thinking_level, and check function calling. Google's guides: for gemini-3.8-flash, for gemini-3.5-flash-lite
Find every place your code uses them
Run this in the root of your repository. No install, account or API key; it lists every OpenAI, Anthropic and Google model ID it finds, with file, line, retirement date and replacement:
npx -y @aidriftline/driftline . A plain search works too:
grep -rnE '"(gemini-2\.5-pro|gemini-2\.5-flash)"' . It misses model names set outside the code: environment variables, deployment settings, values stored in a database, and defaults inside SDKs or agent frameworks you never configured.
Test before 1 December 2026
A new model answers the same prompt differently: format, length, tone, and whether JSON still parses. Take 10 to 20 real prompts from your logs, run each on the old model and the replacement, and compare. Do it while the old model still answers; after the date there is nothing left to compare against.
Driftline Pro does this in your GitHub Actions: it runs the prompts your app sends on the current model and the replacement, and shows what changed in the pull request.
Get a reminder
The free Driftline GitHub Action flags retired and soon-to-retire model IDs on every pull request, so the next retirement doesn't come as a surprise.
Questions
Will calls to gemini-2.5-pro keep working after 1 December 2026?
They keep being answered, by gemini-3.8-flash and gemini-3.5-flash-lite. Google still lists request changes to make before then, see above.
I build Driftline, which tracks these retirement dates. The current list, with one page per model: aidriftline.com/models.
