Gemini 2.5 is unstable: switch to Gemini 3.x

Published: October 23, 2026
Last checked: October 23, 2026
Status: Monitoring. The models are still listed and callable, but success rates vary

Since October 20, 2026, the Gemini 2.5 text models on LaoZhang API have not had enough upstream capacity, and some requests fail with 429 or 5xx. Google has not deprecated these models and LaoZhang API is keeping them listed. For production traffic, switch to the Gemini 3.x models below; only the model field changes.

Affected models and replacements

Affected model	Switch to	Typical use
`gemini-2.5-flash`	`gemini-3.5-flash` or `gemini-3.6-flash`	General chat, code, multimodal understanding
`gemini-2.5-flash-nothinking`	`gemini-3.5-flash-lite`	Low-latency calls; default thinking level `minimal`
`gemini-2.5-flash-lite`	`gemini-3.1-flash-lite` or `gemini-3.5-flash-lite`	Classification, extraction, batch jobs
`gemini-2.5-pro`	`gemini-3.1-pro-preview`	Complex reasoning and long documents

The replacement models are working normally and support both the OpenAI-compatible and the native Gemini API. Check prices and token groups on Models and pricing.

Who is affected

Affected: requests to the four Gemini 2.5 text model IDs above, through either the OpenAI-compatible or the native Gemini API.
Not affected: Gemini 3.x text models and the image model gemini-2.5-flash-image (Nano Banana).

Failed requests return 429 (upstream busy) or 5xx (upstream error). Some requests to the same model may succeed while others fail. That pattern comes from the capacity shortage, not from your request parameters.

How to migrate

Change the model ID to the Gemini 3.x model in the table. The base URL and API key stay the same.
Test with a small share of traffic using your own prompts. Compare thinking behavior, tool calls, structured output, token usage, and latency.
Confirm price and access: check the unit price on Models and pricing, and make sure your token’s group can call the model under Token management.
Check billing in the usage log: confirm the model name, usage, and charges before moving the rest of your traffic.

OpenAI-compatible request (for the native Gemini API, put the new model in the path, for example /v1beta/models/gemini-3.5-flash:generateContent):

export LAOZHANG_API_KEY="your LaoZhang API key"

curl https://api.laozhang.ai/v1/chat/completions \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.5-flash",
    "messages": [
      {"role": "user", "content": "Turn these meeting notes into three action items"}
    ]
  }'

If you must stay on Gemini 2.5 for now, retry 429 and 5xx responses with backoff and lower your concurrency so retries do not pile up.

FAQ

Has Google retired Gemini 2.5?

No. Google’s deprecations page lists no shutdown date for gemini-2.5-flash, gemini-2.5-flash-lite, or gemini-2.5-pro. The failures come from the upstream capacity currently available to LaoZhang API. Google also recommends its latest Gemini models for new projects.

Will LaoZhang API remove Gemini 2.5?

Not for now; the model IDs stay available. We will update this page and post on the announcements page when the status changes.

Does switching to Gemini 3.x change the price?

Yes. Gemini 3.x is priced differently from 2.5. See Models and pricing for current prices and the usage log for what you were charged.

How do I control thinking on Gemini 3.x?

Gemini 3 models take the thinking level as a request parameter, and their defaults differ: Gemini 3.6 Flash defaults to medium, Gemini 3.5 Flash-Lite to minimal. Do not add suffixes such as -thinking or -nothinking yourself; use the model IDs listed in the console. See the Gemini 3.6 Flash and 3.5 Flash-Lite announcement for details.

Why is Gemini 3.8 Flash not on the list?

Gemini 3.8 Flash is available, but its upstream has also been busy at times recently. For steady production traffic, start with the replacements in the table.

Google: Gemini API deprecations and shutdown dates, Latest Gemini models
LaoZhang API status: based on gateway traffic from October 20 to 23, 2026. Current groups, prices, and charges are on Models and pricing and in the usage log.
Related docs: Native Gemini API, OpenAI SDK setup, Check whether a model is available

​Affected models and replacements

​Who is affected

​How to migrate

​FAQ

​Has Google retired Gemini 2.5?

​Will LaoZhang API remove Gemini 2.5?

​Does switching to Gemini 3.x change the price?

​How do I control thinking on Gemini 3.x?

​Why is Gemini 3.8 Flash not on the list?

​Sources and related pages