Google is shutting down Gemini 2.5 Pro and Gemini 2.5 Flash in the Gemini API on 1 December 2026. Both models were marked deprecated in the Gemini API changelog on 2 November 2026 and can no longer be selected in Google AI Studio. Unlike most model retirements, your API calls won’t start failing on that date. Google will automatically route requests for the old model IDs to newer models and bill them at the newer models’ rates.
That sounds convenient, but it means an app that still says gemini-2.5-pro will quietly start getting answers from a different, Flash-class model. Output quality, latency, token usage and cost can all change, and the newer models have breaking API changes. If you have code, scripts or tools that call these models, switch them yourself and test before 1 December.
Gemini 2.5 Shutdown at a Glance
| Model ID | Status | Shutdown | From 1 Dec 2026, requests go to |
|---|---|---|---|
gemini-2.5-pro | Deprecated | 1 Dec 2026 | gemini-3.8-flash, billed at Gemini 3.8 Flash rates |
gemini-2.5-flash | Deprecated | 1 Dec 2026 | gemini-3.5-flash-lite, billed at Gemini 3.5 Flash-Lite rates |
gemini-2.5-flash-lite | Not deprecated | No date announced | Keeps working |
Google’s own wording is that “effective December 1, 2026, all API requests to gemini-2.5-pro will automatically route to gemini-3.8-flash“, and the same applies to gemini-2.5-flash with gemini-3.5-flash-lite. The deprecation notice tells developers to update the model string before that date.
Why the Automatic Rerouting Matters
A reroute keeps your requests from erroring, but it isn’t a like-for-like swap:
- Gemini 2.5 Pro goes to a Flash model, not a Pro model. Requests to
gemini-2.5-proland on Gemini 3.8 Flash. For most tasks that’s a newer and capable model, but if you picked Pro for long, careful reasoning, check that the results still hold up. - Gemini 2.5 Flash goes down a tier, to Flash-Lite. Requests to
gemini-2.5-flashland on Gemini 3.5 Flash-Lite, Google’s cheapest current text model. - Your bill changes. Rerouted calls are billed at the new model’s prices, so token costs will differ from what you budgeted for 2.5.
- Some requests may fail anyway. Google lists breaking API changes in the newer models (below). A request that relies on old parameters can be rejected even after it’s rerouted.
Breaking API Changes to Check
According to the Gemini API changelog, the breaking changes between Gemini 2.5 and the 3.x replacements include:
- Thinking controls:
thinking_levelreplacesthinking_budget. If your code sets a token budget for thinking, change it to a level. - Function calling: stricter function calling and
thought_signaturevalidation. If you build multi-turn tool-calling conversations by hand, send the thought signatures back exactly as you received them. The official SDKs handle this for you when you pass the full response history. - Sampling parameters: some sampling parameters have been removed. Drop any settings the new model rejects.
- Prefilled model turns: you can no longer start the model’s reply for it by ending the conversation with a partial
modelturn.
Google publishes a migration guide for each replacement model. Read the Gemini 3.8 Flash and Gemini 3.5 Flash-Lite guides linked from the Gemini API deprecations page before you switch.
How to Find Gemini 2.5 Model IDs in Your Code
Search your repositories, configuration files and scripts for the two retiring IDs. The pattern below skips gemini-2.5-flash-lite, which isn’t affected:
grep -rnE 'gemini-2\.5-(pro|flash)\b' . | grep -v 'flash-lite'
Also check places that grep in a repository won’t reach: environment variables, .env files, CI/CD variables, secrets managers, and settings in third-party tools (chatbots, IDE plugins, n8n or Zapier workflows, WordPress AI plugins) where you chose a Gemini model by name. Preview IDs such as gemini-2.5-pro-preview-06-05 were already shut down earlier, so if you still see one, it is being served by a newer model today.
How to Migrate
The fix is usually a one-line change to the model string. With the official Google Gen AI SDK for Python, switch the model and replace any thinking budget with a thinking level:
from google import genai
from google.genai import types
client = genai.Client() # reads GEMINI_API_KEY from the environment
response = client.models.generate_content(
model="gemini-3.8-flash", # was: gemini-2.5-pro
contents="Summarise this nginx error log in three bullet points.",
config=types.GenerateContentConfig(
thinking_config=types.ThinkingConfig(thinking_level="low"),
),
)
print(response.text)
If you call the REST API directly, change the model name in the URL:
curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.8-flash:generateContent" \-H "x-goog-api-key: $GEMINI_API_KEY" \-H "Content-Type: application/json" \-d '{"contents":[{"parts":[{"text":"Explain what an inode is"}]}]}'
Before you switch production traffic, run the same set of real prompts through the old and new models and compare the answers, response times and token counts. Google’s defaults point Pro users to Gemini 3.8 Flash and Flash users to Gemini 3.5 Flash-Lite, but you don’t have to accept that pairing. If your Gemini 2.5 Flash workload needs more quality than Flash-Lite gives, move it to gemini-3.8-flash instead. See our Gemini model page for the current lineup, API IDs and context windows.
Note: Put the model ID in one configuration value or environment variable instead of hard-coding it in several places. The next retirement then becomes a one-line change.
Other Gemini 2.5 Models With Shutdown Dates
Gemini 2.5 Flash-Lite stays available with no shutdown date, but several other 2.5 variants are on Google’s list:
| Model ID | Earliest shutdown | Replacement |
|---|---|---|
gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-tts | 18 Dec 2026 | gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts |
gemini-2.5-flash-image | 15 Mar 2027 | gemini-3.1-flash-lite-image |
gemini-2.5-flash-native-audio-preview-12-2025 | 31 Mar 2027 | gemini-3.8-live |
Google notes that the dates on its deprecations page are the earliest possible shutdown dates. These apply to the Gemini API (Google AI Studio keys). Gemini models on Google Cloud’s Vertex AI follow a separate schedule, so check the Vertex AI model versions page if you use Gemini through Google Cloud.
Anthropic and OpenAI are retiring models in the same window, including Claude Sonnet 4.5 on 30 November and GPT-5 and o3 on 11 December. See every date in one calendar in AI model retirements between now and April 2027.
Who Is Affected?
Since 18 September 2026, Google has limited Gemini 2.5 access to users who had already been using these models, and new projects were told to start on Gemini 3.5 Flash-Lite or 3.8 Flash. So this shutdown mainly affects existing apps and integrations built during 2025 and early 2026, which is exactly the code nobody has looked at for a while. If a tool you depend on lets you pick a Gemini model, check that setting too.
If you’d rather not deal with forced migrations at all, open-weight models running on your own hardware are never retired. See how to run models locally with Ollama and use the LLM VRAM calculator to size the GPU.
Sources: Gemini API deprecations and Gemini API changelog (Google, checked 16 November 2026).
How did it make you feel?
