MOUNTAIN VIEW, March 3 — Google released Gemini 3 on Tuesday, restructuring its Gemini AI model lineup around a single trained model family with two consumer tiers, the company said in a blog post and at a press event held at its Mountain View headquarters.
The release replaces the previously fragmented Gemini 2.0 Flash, 2.0 Pro, and 2.0 Experimental lineup with a unified Gemini 3 family available in Pro and Ultra configurations. Both tiers ship with a 2-million-token context window for paid customers — double the largest window Google previously offered — and what the company described as “substantially improved” multimodal reasoning across video, image, audio, and text.
“Gemini 3 is the model the team has been building since the merger of Google Brain and DeepMind,” said a Google spokesperson, referring to the 2023 reorganization that created Google DeepMind. “We have spent the last eighteen months on a substantially redesigned training stack, and Gemini 3 is the first frontier release that comes out of it.”
Pricing and product placement
Gemini 3 Pro is available on the consumer Gemini Advanced subscription at $20 per month, the same price as the previous Gemini 2.0 Advanced tier. Gemini 3 Ultra is priced at $30 per month and includes a higher daily query allowance, priority access during peak hours, and access to the company’s Deep Research and Deep Think features.
For developers, Gemini 3 Pro is priced at $1.25 per million input tokens and $5.00 per million output tokens through the Google AI Studio API, undercutting the OpenAI and Anthropic Opus-tier API pricing. Ultra is priced at $4.00 input and $20.00 output per million tokens.
Native video reasoning
The most-discussed feature in early developer reactions was Gemini 3’s native video reasoning. Google said the Pro and Ultra models can ingest a single video file of up to one hour in length and answer time-stamped questions about its content without intermediate transcription. The company demonstrated the feature at the press event with a 47-minute lecture recording, against which the model returned a chapter outline, a glossary of technical terms, and direct answers to viewer-style questions.
“Native video at hour-length is a real capability shift,” said Beatrix Mendelsohn-Liu, an analyst at Galileo Insights. “Most enterprise video workloads have been blocked on transcription-plus-RAG architectures that lose the visual signal. Gemini 3 Pro is the first frontier model where the video stays the video.”
Competitive context
The release lands roughly two weeks before OpenAI is expected to ship GPT-5 and roughly five weeks after Anthropic shipped Claude Opus 4. All three frontier-model vendors are now in market with a 2026-generation flagship, in what analysts have characterized as the most compressed competitive cycle since the original GPT-4 launch in 2023.
A Google spokesperson said Gemini 3 is available immediately on the Gemini consumer app, Google AI Studio, Vertex AI, and as the default model behind Workspace AI features for enterprise customers.
Asari Whitfield-Asari covers AI tools and developer infrastructure for Consumer Tech Wire.