Google unveils Gemini 3 Flash AI model for faster performance

Google has launched Gemini 3 Flash, a new AI model focused on faster performance and lower costs, now available globally across its products.

Gemini 3 Flash is part of the Gemini 3 AI family and is being rolled out to developers via Gemini API, Google AI Studio, and Google Antigravity, as well as to consumers in the Gemini app and AI Mode in Search.

The model is also accessible to enterprises through Vertex AI and Gemini Enterprise.

According to Google, Gemini 3 Flash delivers improved speed and efficiency, while offering performance on reasoning benchmarks such as GPQA Diamond and Humanity’s Last Exam that is comparable to larger models.

It is priced at US$0.50 per 1 million input tokens and US$3 per 1 million output tokens.

Gemini 3 Flash replaces the previous 2.5 Flash model as the default in the Gemini app and is being gradually introduced as the default for AI Mode in Search worldwide.

🔗 Source: Google

🧠 Food for thought

Implications, context, and why it matters.

Gemini 3 Flash pricing places it between budget and premium tiers

At $0.50 per 1M input tokens plus $3 per 1M output tokens, Gemini 3 Flash costs less than GPT-4o while exceeding GPT-4o mini’s $0.15/$0.60 rates 1.
Google aims at the mid tier with a direct matchup against Claude 3.5 Haiku at ~$0.80/$4 1.
Google has not posted head-to-head results on latency (response time) and context window (how much text the model can process at once) versus GPT-4o mini or Claude 3.5 Haiku. On price alone, Gemini 3 Flash runs about 3–5x higher than GPT-4o mini per token while undercutting Claude 3.5 Haiku 1.
Google replaced 2.5 Flash as the default model, which signals confidence 2 3. Teams running production AI apps should watch deprecation schedules and model lifecycle guidance to plan migrations 2 3.

Third-party evaluation and migration tooling providers can capture value during the 2.5 to 3 Flash transition

With Gemini 3 Flash as the new default and 2.5 Flash likely to retire based on past practice, teams with production workloads need to check that the new model holds or lifts performance on their use cases 2 3.
Independent AI shops or DevOps platforms can provide regression test suites, A/B routing, plus cost checks that compare Gemini 3 Flash with its predecessor or small rivals like GPT-4o mini 1.
Early movers who build full evaluation frameworks before any formal retirement dates can become the default pick for migrations 2 3.