Gemini 3 Flash

Claims and evidence

• Released 17 December 2025Released as a preview model for developers and enterprises; rollout in the Gemini app and AI Mode in Search started the same day. ● Primary source[1] [4]December 17, 2025 [3]gemini-3-flash-preview: release date [5]Versions: Release date

• Status PreviewStill a preview model in the Gemini API and on Gemini Enterprise Agent Platform on 2026-10-01; no shutdown date announced. ● Primary source[2]Gemini 3, Preview [3]Gemini 3 models, Preview models: gemini-3-flash-preview [5]Versions: Launch stage
• Replaced by Gemini 3.6 FlashRecommended replacement as listed on the Gemini deprecations page when checked on 2026-10-01; no shutdown date announced. ● Primary source[3]gemini-3-flash-preview: recommended replacement

• Successor of Gemini 2.5 FlashThe announcement says Gemini 3 Flash replaces 2.5 Flash as the default model in the Gemini app and builds on the Flash series. ● Primary source[1]For everyone: Gemini 3 Flash is rolling out globally
• Derived from (other) Gemini 3 ProThe model card says Gemini 3 Flash is based on Gemini 3 Pro and built off its reasoning foundation; it does not name the method. ● Primary source[6]Model Information: Model dependencies

• Change · Availability Replaced Gemini 2.5 Flash as the default model in the Gemini app and began rolling out as the default model for AI Mode in Search.Compared with Gemini 2.5 Flash ● Primary source[1]
• Change · Architecture Based on Gemini 3 Pro, according to Google's model card, which says it builds on that model's reasoning foundation.Compared with Gemini 2.5 Flash ● Primary source[6]Model Information: Model dependencies
• Change · Tool use Supports the Computer Use tool in the Gemini API, added on 29 January 2026.Compared with Gemini 2.5 Flash ● Primary source[4]January 29, 2026 [7]Capabilities: Computer use

• Input text, image, audio, videoPDF input is also supported. ● Primary source[7]Supported data types: Inputs [6]Inputs
• Output text ● Primary source[7]Supported data types: Output
• Feature Reasoning modeThinking levels set the balance between quality, cost and latency. ● Primary source[7]Capabilities: Thinking [6]Description
• Feature Function calling ● Primary source[7]Capabilities: Function calling
• Feature Computer useComputer Use tool support added on 2026-01-29. ● Primary source[4]January 29, 2026 [7]Capabilities: Computer use
• Context window 1,048,576 tokensInput token limit; the output limit is 65,536 tokens. The model card gives a context window of up to 1M. ● Primary source[7]Token limits: Input token limit
• Parameters Not disclosedNot disclosed by Google. No claim madeNo source
• Access API, consumer appAPI: the Gemini API (Google AI Studio) and Vertex AI, in preview. Consumer: the Gemini app and AI Mode in Search. ● Primary source[1]Try Gemini 3 Flash today

Lineage

Variants

No variants recorded in this record.

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from Google on AI Radar →