Gemini 3 Pro: Shut Down by Google on 9 March 2026

This model has been shut down

Gemini 3 Pro is no longer available. Google's own documentation carries the notice plainly: Gemini 3 Pro Preview is deprecated and has been shut down as of 9 March 2026, with a recommendation to migrate to Gemini 3.1 Pro Preview to avoid service disruption. The identifier Google's notice names is the preview endpoint, gemini-3-pro-preview. That is the identifier the shutdown applies to, and it is the page this entry links to.

Why the page is still here

A retired model is not deleted from this roster, and that is deliberate. Its URL is linked from comparison pages, from anywhere else on the web that picked it up, and from documentation and code that outlived the model itself — and a reader arriving from one of those needs to be told what happened, not handed a 404. The retirement is recorded in the specification table as well as in this text.

What it was, while it ran

Gemini 3 Pro was the Pro-tier model of Google's Gemini 3 generation, carrying a 1 million token context window and listed at $2 per million input tokens and $12 per million output. On LMArena's text leaderboard Gemini 3 Pro has 1485 Elo (rank 21); by category, 1517 Elo in Coding (rank 44), 1477 Elo in Math (rank 45), 1499 Elo in Expert (rank 45) and 1485 Elo in Creative Writing (rank 11). Google's Gemini 3 developer guide describes the generation as built on state-of-the-art reasoning and aimed at agentic workflows, autonomous coding and complex multimodal work, and introduced a thinking_level parameter giving developers direct control over how deeply the model reasons before answering.

Successor

Google's notice names Gemini 3.1 Pro Preview as the migration target. On LMArena, Gemini 3.1 Pro has 1487 Elo (rank 20). The Gemini 3 developer guide covers the generation's API surface, including thinking_level, which continues into the newer models.

Specification while it ran

Google's model page lists gemini-3-pro-preview with text, image, video, audio and PDF input, text output, an input limit of 1,048,576 tokens and an output limit of 65,536 tokens. Supported capabilities were caching, code execution, file search, function calling, Google Search grounding, structured outputs, thinking and URL context; audio generation, computer use, grounding with Google Maps, image generation and the Live API were not supported. It could be used through the Batch API but not with Flex or Priority inference. The page records its latest update as November 2025.

The Gemini 3 thinking controls

The Gemini 3 developer guide states that the family uses dynamic thinking by default and introduced the thinking_level parameter, which sets the maximum depth of the model's internal reasoning before it answers; Google describe the levels as relative allocations rather than strict token guarantees. When thinking_level is not specified, Gemini 3 uses high. On Gemini 3.1 Pro the minimal level is not supported, while low, medium and high are. Google list Gemini 3.1 Pro at $2 input and $12 output per million tokens for prompts up to 200,000 tokens, and $4 and $18 above that, with a January 2025 knowledge cutoff. Google have since marked the Gemini 3 guide as deprecated in favour of the Gemini 3.5 Flash developer guide.

Official Resources