Gemini 3 is a model family—not a single current chatbot option

By Chatbot Daily · Updated November 1, 2026

Gemini 3 is officially released, but the original Gemini 3 Pro text model has been replaced in the API. Google launched the family on November 18, 2025. The original Pro Preview model was shut down on March 9, 2026; Google says its old API name now points to Gemini 3.1 Pro Preview. Later 3-series models remain available. Google’s release notes

Gemini 3 is also no longer Google’s newest model generation: Gemini 4 Argon became generally available in the API on October 13, 2026. That does not mean every Gemini 3 model has been retired. Official release record

Which Gemini 3 model is available now?

These are the main text-output options to distinguish in the Gemini Developer API, not an exhaustive catalog. The consumer app’s plan labels are covered below.

Model	API model ID	Current status and purpose
Original Gemini 3 Pro	gemini-3-pro-preview	Original model shut down; the name points to 3.1 Pro Preview. Do not use it to reproduce original 3 Pro behavior. Release notes
Gemini 3.1 Pro	gemini-3.1-pro-preview	Still a preview model; the Pro replacement for complex reasoning and multimodal work. Model documentation
Original Gemini 3 Flash	gemini-3-flash-preview	Still listed in preview, with no shutdown date announced. Google’s replacement recommendation is 3.6 Flash. Lifecycle schedule
Gemini 3.6 Flash	gemini-3.6-flash	Stable and still supported; an alternative to 3.8 Flash for workflows that do not need its more iterative reasoning. 3.8 guidance · GA release record
Gemini 3.8 Flash	gemini-3.8-flash	Stable; designed for long-running coding and agent workflows. Google describes it as its most intelligent Flash model. Model documentation
Gemini 3.5 Flash-Lite	gemini-3.5-flash-lite	Stable; optimized for low-cost, high-volume tasks such as document parsing and simple extraction. Model documentation

For a new 3-series application, evaluate a stable model first rather than copying an original preview ID from a launch tutorial. Choose Flash-Lite when cost and latency are the main constraints; compare 3.8 Flash and 3.1 Pro when the task requires more complex reasoning.

Who gets access in the Gemini app?

For personal accounts, Google’s current subscription page advertises the following access. These are app entitlements, not an API model-ID list. In particular, its generic Flash and Flash-Lite labels do not establish which numbered API version you receive. Gemini subscriptions

Personal-account plan	Advertised model access	App context window
Free	Flash-Lite	32K tokens
Google AI Plus	Flash	128K tokens
Google AI Pro	Pro and Flash, plus Gemini 4 Argon	1 million tokens
Google AI Ultra	Highest access to Pro and Flash, plus Gemini 4 Argon	1 million tokens

The subscription page also identifies 3.1 Pro among its paid-plan benefits. The context-window figures come from Google’s app limits documentation. A million-token API specification does not give a free app account the same reading capacity.

App usage is now compute-based: prompt complexity, the model or feature used, and conversation length affect consumption. Limits refresh every five hours until the weekly limit is reached. To see your allowance on the web, open Settings → Usage limits. Google says access and limits can change with availability and experimentation. Usage rules

Before upgrading, check country and age eligibility in Google’s help page and the subscription conditions shown for your account. Personal-account rules do not establish access for work or school accounts. Google also notes that mobile and web features can differ. Eligibility guidance · Subscription conditions

Developer access and cost

Google AI Studio and the Gemini Developer API have their own supported-region list. Check that list separately from consumer subscription eligibility.

Gemini 3.1 Pro Preview can be tried at no cost in Google AI Studio, but has no free API tier. Standard API pricing is US$2 input / $12 output per million tokens for prompts up to 200K tokens, rising to $4 / $18 above that threshold. Output pricing includes thinking tokens. Developer guide · Official pricing

Gemini 3.8 Flash has a free API tier and introductory standard paid rates of US$0.75 input / $3.75 output per million tokens through December 31, 2026. Google lists rates of $1.50 / $7.50 from January 1, 2027. Output charges include thinking tokens; tool and caching charges can add to the bill. Official pricing · 3.8 Flash pricing terms

What can Gemini 3 do—and what requires another model?

Gemini 3.1 Pro and 3.8 Flash accept text, images, video, audio and PDFs, but their native output is text. Both list an input limit of 1,048,576 tokens and an output limit of 65,536 tokens. They support thinking, function calling, structured outputs, code execution and search grounding. These are supported capabilities, not evidence that every request automatically searches or runs code. 3.1 Pro specifications · 3.8 Flash specifications

Neither endpoint natively generates images or audio, or supports the Live API. Image generation, speech generation and real-time voice use separate models. For example, Nano Banana Pro is Gemini 3 Pro Image—not the ordinary Pro text endpoint. The official model catalog separates these families.

What changed, and what should you retest?

Google positioned the original Gemini 3 release as an improvement over 2.5 in reasoning, multimodal understanding and coding. It subsequently introduced 3.1 Pro as a further core-reasoning upgrade. Those are Google’s published assessments, not a guarantee of better results on your particular documents or codebase. Gemini 3 launch · 3.1 Pro announcement

The concrete migration changes matter more than the family number:

• Reasoning controls differ by version. Gemini 3 uses thinking_level, but supported values and defaults vary. Original 3 Flash supports minimal; 3.8 Flash rejects it and defaults to medium. Do not carry settings across unchanged. 3-series controls · 3.8 migration guidance
• Document and video token use can change. Google’s 2.5 migration guidance warns that Gemini 3 defaults may increase PDF token consumption while decreasing video consumption. Retest dense PDFs for accuracy, context usage and cost. Migration guidance
• Multi-turn integrations need reasoning context. Preserve thought signatures when managing history manually; the stateful Interactions API manages them for you when using previous_interaction_id. Developer guide
• Later models have additional API changes. The 3.8 guide tells developers to remove deprecated sampling parameters, unsupported candidate_count and prefilled model turns. Its more iterative reasoning can also consume more tokens on complex tasks. 3.8 migration guidance

Before switching a working integration, run representative tasks against an explicit model ID and compare correctness, latency, token consumption and tool behavior. Stop the migration if required output formats or tool calls break, or if cost exceeds your budget. For the broader access-checking method, see how to read an AI tool changelog.