XNews

Google releases Gemini 3 Flash

Google releases Gemini 3 Flash

Google releases Gemini 3 Flash as a speed-optimized model. - Core positioning: 'Frontier Intelligence Built for Speed'. - Not a lightweight: inherits Gemini 3 series reasoning power. - Performance exceeds Gemini 2.5 Pro in key scenarios. - Token efficiency: ~30% fewer tokens than Gemini 2.5 Pro. - Latency: reported 3x faster on everyday tasks. - Pricing: input $0.50/1M tokens; output $3.00/1M tokens. - Benchmarks: GPQA Diamond 90.4%; MMMU Pro 81.2%. - Code score: SWE-bench Verified 78%, surpassing Gemini 2.5 series. - Availability: default in Gemini App; integrated into Google Search (AI Mode). - Multimodal: real-time video, image analysis, drawing and voice-driven prototyping. - Developer access: Google AI Studio, Vertex AI, Google Antigravity agent platform. - Target use cases: real-time video analysis, in-game assistants, A/B test design. - Production users: JetBrains and Figma reported using it. - Demo: public video compares Gemini 3 Flash vs Gemini 3 Pro on Pokemon Crystal gameplay.

Key insights

Gemini 3 Flash emphasizes speed while retaining reasoning

Gemini 3 Flash is presented as 'Frontier Intelligence Built for Speed', combining the Gemini 3 series' reasoning capabilities with lower latency and lower token consumption. The model is positioned not as a minimal lightweight but as a faster variant that keeps robust inference strength. Documentation reports average token usage reductions of about 30% versus Gemini 2.5 Pro and execution speeds roughly three times faster on routine tasks, indicating an explicit trade toward efficiency without sacrificing core intelligence.

Availability across consumer and developer platforms

Google has deployed Gemini 3 Flash broadly: it is the default model in the Gemini App and supports Google Search's AI Mode for concise, real-time answers. The model supports multimodal real-time analysis (video, images, drawing interpretation, voice-driven prototyping) for general users. For developers and enterprises, Gemini 3 Flash is accessible via Google AI Studio, Vertex AI, and the new agent development platform Google Antigravity. Google cites production use by companies such as JetBrains and Figma for low-latency, high-throughput scenarios.

Benchmarks, code capabilities, pricing and demonstrations

Benchmark results show Gemini 3 Flash achieving 90.4% on GPQA Diamond and 81.2% on MMMU Pro; code capability is reported at 78% on SWE-bench Verified, outperforming Gemini 2.5 series and nearing larger models. Pricing published for the model lists $0.50 per million input tokens and $3.00 per million output tokens. Google claims the model outperforms Gemini 2.5 Pro in many cases and uses ~30% fewer tokens. A user-shared video demonstrates Gemini 3 Flash versus Gemini 3 Pro playing Pokemon Crystal, serving as a public performance comparison.

Source posts