If by "Gemini 2.5 Flash-Lite Preview" you mean gemini-2.5-flash-lite-preview-09-2025, then yes: Google's Gemini API deprecations page lists that exact preview endpoint for shutdown on March 31, 2026, and the official replacement is gemini-3.1-flash-lite-preview. But that is not the same thing as stable gemini-2.5-flash-lite disappearing today. Google lists the stable model separately, and its current shutdown date is July 22, 2026. That distinction is the whole story behind this keyword.
The confusion is understandable because Google now has three overlapping Flash-Lite lifecycle layers in public docs: an older 06-17 preview that already died, a newer 09-2025 preview that is now scheduled to shut down, and the live stable gemini-2.5-flash-lite lane that still exists. On top of that, the public Vertex AI model page still shows the 09-2025 preview version, which makes some users think the deprecation notice must be partial or mistaken. It is not that simple. The real question is which exact endpoint your code or prompt surface is using, and whether you are on the Gemini API / Google AI Studio lifecycle path or a different product surface.
TL;DR
gemini-2.5-flash-lite-preview-09-2025is the endpoint on the March 31, 2026 shutdown clock.- stable
gemini-2.5-flash-liteis separate and currently has a July 22, 2026 shutdown date in Google's lifecycle docs. - the official successor is
gemini-3.1-flash-lite-preview, but it is much more expensive, so do not treat the migration as a free rename.
If you only need the practical answer, use the table below and then keep reading only the section that matches your setup.
| What you are using now | Current status in public docs | Date you should care about | Best next move |
|---|---|---|---|
gemini-2.5-flash-lite-preview-09-2025 | Deprecated preview | March 31, 2026 | Plan migration now; Google's replacement is gemini-3.1-flash-lite-preview |
gemini-2.5-flash-lite | Stable | July 22, 2026 | No emergency rename today; benchmark and plan a cost-aware migration |
gemini-3.1-flash-lite-preview | Current preview successor | No shutdown date announced in Gemini API docs | Use this if you want the documented successor lane |
| Vertex AI page still shows preview version | Surface-specific ambiguity | Check your actual surface and endpoint behavior | Confirm before mass-renaming, but do not ignore the Gemini API deprecation signal |
gemini-flash-lite-latest or hidden wrapper aliases | Alias risk | Depends on what the alias resolves to | Inspect the resolved model string before assuming you are safe |
The key point is that the discontinuation story is about an endpoint, not a family name. If your code, prompt tool, or saved config only says "Flash-Lite," you are still missing the part that determines whether this is urgent or not.
Is Gemini 2.5 Flash-Lite Preview actually being discontinued?
For the Gemini API, the cleanest official answer is yes: gemini-2.5-flash-lite-preview-09-2025 is scheduled for shutdown on March 31, 2026. Google says that directly on the public Gemini deprecations page, and the same table names gemini-3.1-flash-lite-preview as the recommended replacement.
The reason people still doubt the answer is that Google's public docs do not present this as one tidy narrative. The changelog shows the lifecycle in pieces:
gemini-2.5-flash-lite-preview-06-17appeared first in June 2025- stable
gemini-2.5-flash-litearrived in July 2025 gemini-2.5-flash-lite-preview-09-2025arrived later in September 2025- in March 2026, Google announced that the 09-2025 preview would be shut down on March 31
That means the phrase "Gemini 2.5 Flash-Lite Preview" can refer to more than one historical model ID. If someone remembers an old preview alias, an AI Studio model picker, or a community post from a different month, they may be remembering the right family name but the wrong endpoint. That is why many searchers still feel unsure even after opening the official docs.
The other source of confusion is product scope. Public Gemini API docs are explicit about the March 31, 2026 shutdown date for the 09-2025 preview endpoint. But the public Vertex AI Flash-Lite page still shows a preview version block for gemini-2.5-flash-lite-preview-09-2025 alongside GA gemini-2.5-flash-lite. So the safe wording is not "Google killed Flash-Lite Preview everywhere." The safe wording is:
Google's Gemini API lifecycle docs list gemini-2.5-flash-lite-preview-09-2025 for shutdown on March 31, 2026. Stable gemini-2.5-flash-lite is separate. Public Vertex docs still show the preview version, so you should verify your actual product surface before changing production routing.
That is a more annoying answer than a clean yes-or-no, but it is the one the public sources support.
The confusing part: preview endpoint, stable model, and older preview IDs are different things
Most weak pages around this keyword fail because they answer the family name and skip the endpoint split. That leads to advice that sounds dramatic but is operationally useless. You need to separate three different things.
First, there is stable gemini-2.5-flash-lite. Google's dedicated model page describes it as the current cost-efficient multimodal model in the 2.5 family and marks it as Stable. The same page marks gemini-2.5-flash-lite-preview-09-2025 as Deprecated. That single Versions block is enough to disprove the common panic interpretation that "Flash-Lite Preview discontinued" means "Flash-Lite is already dead."
Second, there is the deprecated 09-2025 preview endpoint. This is the model string that matters for the March 31, 2026 deadline. If this exact ID appears in your code, config, saved prompt surface, or wrapper defaults, you should treat the migration as urgent.
Third, there is the older 06-17 preview endpoint. That one is already old history. Google's changelog shows it in the November 4, 2025 deprecation notice with a November 18, 2025 shutdown date. That historical detail matters because it explains why community posts, copied code, and stale screenshots can talk about "Flash-Lite Preview" but still be pointing at a different lifecycle event than the one you are facing now.
This is also why blindly searching your repo for "Flash-Lite" is not enough. Search for the full model strings:
bashrg "gemini-2.5-flash-lite-preview-09-2025|gemini-2.5-flash-lite|gemini-flash-lite-latest"
If you only see gemini-2.5-flash-lite, you are not in the March 31 emergency bucket. If you see gemini-2.5-flash-lite-preview-09-2025, you are. If you see an alias like gemini-flash-lite-latest, you need to figure out what it resolves to now rather than assuming the alias is doing what you think.
That last point is more important than it sounds. Teams often migrate the obvious application code and forget:
.envfiles- internal admin tools
- saved prompt templates
- notebooks
- evaluation harnesses
- batch jobs
- wrapper defaults
Deprecation bugs are rarely caused by the main code path alone. They usually survive in one forgotten configuration surface.
What you should switch to right now
The official replacement for gemini-2.5-flash-lite-preview-09-2025 is gemini-3.1-flash-lite-preview. That is the answer Google gives on the deprecations page, and if your goal is to stay on the documented successor path, that is the model ID you should test first.
But that does not automatically mean every user should treat the migration as a free rename. Two practical questions matter:
- Are you actually on the deprecated preview endpoint now?
- Do you want the official successor lane, or do you mainly want to stay on the cheapest live 2.5 Flash-Lite economics?
If you are on gemini-2.5-flash-lite-preview-09-2025 inside the Gemini API or Google AI Studio, the cleanest official move is:
textgemini-2.5-flash-lite-preview-09-2025 -> gemini-3.1-flash-lite-preview
If instead your main business goal is "keep the low-cost Flash-Lite lane alive as long as possible," then stable gemini-2.5-flash-lite is still a real line, still visible on the official model page, and still priced like the deprecated preview row. That is not the documented replacement Google lists for the preview deprecation, so I would not call it the official migration answer. But it is a valid operational consideration for teams that care more about price discipline and stable status than about adopting the 3.1 successor immediately.
That distinction matches the broader comparison we already published in our Gemini 3.1 Flash-Lite vs Gemini 2.5 Flash-Lite guide. The short version is:
- use
gemini-3.1-flash-lite-previewwhen you want the documented successor lane - stay on
gemini-2.5-flash-litewhen the lowest-cost stable lane still matters more than successor positioning
What you should not do is migrate blindly to 3.1 and then act surprised when the bill changes. The replacement is not priced like the model many teams are leaving behind.
What changes in price, limits, and migration risk
This is where the migration gets real.
On the official pricing page, stable gemini-2.5-flash-lite and deprecated gemini-2.5-flash-lite-preview-09-2025 currently share the same published rates. The official replacement does not.
| Model | Status | Standard input price | Standard output price | Practical takeaway |
|---|---|---|---|---|
gemini-2.5-flash-lite-preview-09-2025 | Deprecated preview | $0.10 / 1M | $0.40 / 1M | Cheap, but now on a March 31 shutdown clock |
gemini-2.5-flash-lite | Stable | $0.10 / 1M | $0.40 / 1M | Same published economics, later July 22 shutdown date |
gemini-3.1-flash-lite-preview | Current preview successor | $0.25 / 1M | $1.50 / 1M | Official replacement, but materially more expensive |
That means the recommended replacement is:
- 2.5x more expensive on input
- 3.75x more expensive on output
This is the piece most current result pages bury. Users do not just want to know whether a model is disappearing. They want to know whether the migration will quietly blow up the economics of a lane that was chosen specifically because it was cheap.
The rate-limit story is more nuanced. The official rate-limits page says preview models may have more restrictive limits and tells you to inspect your active limits in AI Studio. At the same time, the public Batch API table currently shows the same published batch-enqueued-token ceilings for Gemini 2.5 Flash-Lite and Gemini 2.5 Flash-Lite Preview. So the safe conclusion is not "3.1 definitely gives you better public throughput." The safer conclusion is:
- the public docs do not currently give you a clean throughput win to justify the migration on their own
- preview-limit behavior may still be less predictable than stable behavior
- the replacement case is primarily about lifecycle and model quality, not a documented batch-capacity advantage
That is why the right migration question is not "what is the new name?" It is "does the 3.1 lane create enough value to justify a much higher token bill before I am forced off the old preview line?"
If you need the bigger surrounding quota picture, our Gemini API rate-limits-per-tier guide and Gemini API pricing 2026 guide are the next pages to open.
Troubleshooting after the rename: Gemini API, Google AI Studio, and Vertex AI
This is the section current SERP pages most often skip. Many users do rename the model and still end up thinking the migration failed. Usually the reason is not that the deprecation notice was false. It is that the stale reference lived in a different surface than the obvious one.
Gemini API
If your application calls the Gemini Developer API with the exact model string gemini-2.5-flash-lite-preview-09-2025, treat this as an active migration item. Google's public Gemini API lifecycle docs are explicit enough that waiting until the end of March is unnecessary risk. Update the model string deliberately and re-test with the smallest possible request before you start debugging anything else.
Google AI Studio
If an old prompt, template, or workspace still references the deprecated preview ID, you are dealing with essentially the same problem as API code: a stale model reference. The difference is just where it lives. In practice, AI Studio migrations often fail because teams update code but forget:
- saved prompt workspaces
- internal screenshots
- onboarding docs
- copied notebooks
So if an AI Studio user says "it still worked for me last week," do not immediately conclude the deprecation notice was wrong. Check whether they are using a saved prompt, a different project, or a model alias path that nobody has audited.
Vertex AI
This is the hardest surface to explain cleanly because the public docs create ambiguity. The public Vertex AI page for Flash-Lite still shows stable gemini-2.5-flash-lite as GA and also includes a preview-version block for gemini-2.5-flash-lite-preview-09-2025. That means you should not write a blanket internal announcement that says "Flash-Lite Preview is gone everywhere." But you also should not ignore the Gemini API lifecycle signal just because the public Vertex page still exposes the preview row.
The safest operational rule is:
- if you are using the Gemini Developer API or Google AI Studio, trust the Gemini API lifecycle pages for migration timing
- if you are using Vertex AI, verify the exact endpoint behavior in your project before mass-renaming, but still plan for the preview line to age out rather than assuming it will remain available indefinitely
That is not a satisfying answer, but it is better than pretending the public sources line up more neatly than they do.
Safe migration checklist
If this keyword is coming from a real production workflow, use this order:
- Search for the full deprecated model ID, not just the family name.
- Separate preview
gemini-2.5-flash-lite-preview-09-2025from stablegemini-2.5-flash-liteeverywhere you find it. - Decide whether your primary target is the official successor
gemini-3.1-flash-lite-previewor the cheaper stable 2.5 line. - Re-test with one minimal request before you debug larger prompts or pipelines.
- Audit saved prompts, wrappers, notebooks, and deployment config for stale model strings.
- If you are on Vertex AI, confirm the actual behavior in your surface before bulk changes.
- Re-run your cost model before moving a high-volume lane onto 3.1 Flash-Lite Preview.
If you want one practical default, here it is:
- migrate now if you are explicitly pinned to
gemini-2.5-flash-lite-preview-09-2025 - do not panic if you are already on stable
gemini-2.5-flash-lite - do not treat 3.1 as a free rename because it is not priced that way
If your migration turns into broader 400, 403, or 429 issues after the rename, the right follow-up is our Gemini API error troubleshooting guide, not more guessing about the lifecycle notice.
FAQ
Is stable gemini-2.5-flash-lite already discontinued?
No. As of March 21, 2026, Google's public Gemini API docs still list stable gemini-2.5-flash-lite separately from the deprecated preview row. The current shutdown date Google shows for the stable line is July 22, 2026, not March 31, 2026.
What exactly is shutting down on March 31, 2026?
The official Gemini API deprecations page lists gemini-2.5-flash-lite-preview-09-2025 for shutdown on March 31, 2026. That is the preview endpoint most readers mean when they search this keyword. The same page lists gemini-3.1-flash-lite-preview as the replacement.
Does this affect Vertex AI too?
Public Vertex AI docs still show the preview version, so you should not assume one universal state across all Google surfaces. The safe interpretation is that the Gemini API lifecycle docs clearly signal the preview shutdown, while Vertex behavior should be confirmed in the actual surface you use. That surface split is also why people remain confused even though the public docs exist.