Gemini 3 Flash release notes / what's new

Google DeepMind has released Gemini 3 Flash, a model designed to provide frontier-grade intelligence with high speed and efficiency. Gemini 3 Flash integrates the reasoning capabilities of the Gemini 3 Pro model into a lightweight architecture, making next-generation AI accessible across Google products and developer platforms at a fraction of the cost.

Frontier Reasoning and Multimodal Performance

Gemini 3 Flash delivers performance on PhD-level reasoning and knowledge benchmarks that rivals larger frontier models and outperforms Gemini 2.5 Pro. Key benchmark results include:

  • GPQA Diamond: 90.4%
  • Humanity’s Last Exam: 33.7% (without tools)
  • MMMU Pro: 81.2% (comparable to Gemini 3 Pro)

Beyond raw benchmarks, the model is designed for efficiency. When operating at the highest thinking level, Gemini 3 Flash can modulate its processing time based on complexity. On typical traffic, it uses an average of 30% fewer tokens than Gemini 2.5 Pro to complete everyday tasks with higher performance.

Speed, Cost, and Efficiency

Gemini 3 Flash is optimized for low-latency applications, performing 3x faster than Gemini 2.5 Pro according to Artificial Analysis benchmarking. This speed makes it particularly suitable for agentic workflows and responsive interactive applications.

Pricing Structure:

  • Input Tokens: $0.50 per 1 million tokens
  • Output Tokens: $3.00 per 1 million tokens
  • Audio Input: $1.00 per 1 million tokens

Developer Capabilities and Agentic Coding

Gemini 3 Flash is optimized for iterative development and high-frequency workflows. It demonstrates significant strength in coding agent capabilities, achieving a score of 78% on SWE-bench Verified, which outperforms both the Gemini 2.5 series and Gemini 3 Pro.

These capabilities enable developers to build more intelligent applications, such as in-game assistants or A/B test experiments, that require a combination of deep reasoning and rapid response times. The model is specifically targeted at complex video analysis, data extraction, and visual Q&A.

Global Integration and Consumer Access

Gemini 3 Flash is being integrated across the Google ecosystem to replace older models and enhance user experiences:

  • Gemini App: Gemini 3 Flash is now the default model, replacing Gemini 2.5 Flash. This allows users to transform unstructured voice thoughts into functioning app prototypes or convert multimodal content (video and images) into actionable plans.
  • AI Mode in Search: The model is the new default for AI Mode in Search, improving the parsing of nuanced queries and providing comprehensive, visually digestible responses that integrate real-time local information and web links.

Availability

Gemini 3 Flash is available in preview through the following channels:

  • Developer Tools: Gemini API in Google AI Studio, Google Antigravity, Vertex AI, Gemini Enterprise, Gemini CLI, and Android Studio.
  • Consumer Products: The Gemini app and AI Mode in Search.

Sources