Google Gemini 3.8 Flash Rapid Evolution, Complex Version Management

Gemini 3 Series Release Timeline (Nov 2025 – Sep 2026)

This table lists only the text and live-voice model releases from Google’s Gemini API release notes. Image, music, video, and text-to-speech models were released separately.

Gemini 3.8 Flash and the Models Around It

• Long input: Gemini 3.8 Flash accepts up to 1,048,576 input tokens (about 1 million) and returns up to 65,536 output tokens. It takes text, images, video, audio, and PDF as input and answers in text.
• Real-time voice (Live API): Gemini 3.8 Live, generally available since September 15, supports real-time voice conversations with reduced latency. Gemini 3.8 Live Extended Thinking, which reasons in the background during a conversation, is a separate model.
• Flash-focused releases: According to the release notes, every text model that reached general availability in 2026 was a Flash or Flash-Lite model. In the public API, the newest Pro model is still 3.1 Pro Preview from February, as of October 6. Google announced its next-generation model, Gemini 4 Argon, on September 30, but for now it is offered only to selected early users, such as some members of Fairwind, Google’s cybersecurity partner program. Google says paid API customers and Google AI Ultra subscribers come next, but has not given a date. Our Gemini 4 Argon analysis covers the details.
• Image generation and editing: Gemini 3.1 Flash Image and 3 Pro Image became generally available on May 28. 3.1 Flash Image can turn a video into a still such as a thumbnail or poster, but its output is always an image. Video generation is handled by separate models such as Veo and Gemini Omni Flash. On October 6, Google released Nano Banana 2.1 (gemini-nano-banana-2.1), the successor to 3.1 Flash Image, and announced that 3.1 Flash Image will shut down on October 29.

When You Actually Have to Change Your Code

The checklist is short. First, find the exact model ID in your configuration and check whether it is a latest alias: an alias such as gemini-flash-latest can start pointing to a different model without any code change. Next, read the shutdown notices and migration guides for changes to request format, tools, and sampling settings. Finally, test quality and cost on the same tasks before anything reaches your live service. Cost can change even when your code does not. Gemini 3.6, 3.7, and 3.8 Flash are on an introductory rate of $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026; on January 1, 2027, the price doubles to $1.50 and $7.50.

With versions arriving this often, it is easy to lose track of which model to pick. But older models do not stop working the moment a new one ships. As of October 6, Google’s deprecations page lists no shutdown date for the GA models 3.5, 3.6, 3.7, and 3.8 Flash. Among GA 3.x text models, only 3.1 Flash-Lite has a scheduled shutdown (no earlier than May 7, 2027).

That does not make GA a lifetime guarantee. The image model 3.1 Flash Image got its shutdown date about five months after going GA, with a little over three weeks between the notice and the switch-off.

In 2026, the model IDs that actually had to be replaced were mostly previews and older families. Gemini 3 Pro Preview shut down on March 9, and its model ID now points to 3.1 Pro Preview. 3.1 Flash-Lite Preview shut down on May 25, and the Gemini 2.0 Flash family on June 1.

“Deprecated” does not mean “already gone.” In Google’s documentation, a deprecation means the model is no longer supported and will be shut down in the near future; the shutdown is when the endpoint is actually turned off. The July 21 release notes marked the sampling parameters temperature, top_p, and top_k as deprecated, and the 3.8 Flash migration guide tells developers to remove them. The release notes do not give a date when older models will start ignoring or rejecting them.

Restricted access is not a shutdown either. Since September 18, Google has limited the 2.5 models to users who have actively used them in the past, but it says they are not deprecated and will keep being served through the API until further notice. For new projects, Google recommends 3.5 Flash-Lite or 3.8 Flash.

What to Weigh Before You Adopt

• Faster releases are not automatically better. If you run a production service, model stability and backward compatibility often matter more than the newest features.
• Gemini 3.8 Flash stands out for long input (about 1 million tokens) and mixed-format input. On the Terminal-Bench 4.0 and FrontierCode 1.1 scores compared on this blog, however, it scored below Claude Fable 5.1 and GPT-6 Astra.
• Keep model IDs in one configuration file instead of scattering them through your code, so you can find what to change quickly when a shutdown notice arrives. Notices can come with little lead time, so check the release notes and the deprecations page regularly.