Google Gemini 3.5 Flash + Omni multimodal/agentic; new Flash checkpoint on LM Arena (likely 3.6 or 4)
Key Questions
What improvements does Gemini 3.5 Flash offer over previous versions?
Gemini 3.5 Flash is now the default in the Gemini app and Google Search AI Mode, achieving 83.6% on agentic MCP Atlas while delivering 4x faster output than prior models. It includes native computer use capabilities and adversarial training, though coding performance still trails competitors.
What is the new Gemini Omni model and its key features?
Gemini Omni is a new video model supporting conversational editing and physics simulation. It is available as Gemini Omni Flash for interactive video generation at $0.10 per second.
What new image generation models has Google released?
Google introduced Nano Banana 2 Lite, a fast and low-cost image generation model that produces images in about 3 seconds for $0.03 each. It also offers Gemini 3 Pro Image (Nano Banana Pro) with improved throughput.
What does the new Gemini Flash checkpoint on LM Arena indicate?
A new anonymous Gemini Flash checkpoint appeared on LM Arena, likely representing version 3.6 or 4 as an incremental update over 3.5 Flash. Early testers note improvements but no benchmarks are available yet, and the Pro version remains delayed.
How does Gemini 3.5 Live Translate work?
Gemini 3.5 Live Translate provides streaming speech-to-speech translation across more than 70 languages. It integrates multimodal capabilities including the new Gemini Embedding 2 for SOTA embeddings.
Gemini 3.5 Flash now default in Gemini app and Google Search AI Mode, surpassing GPT-5.5 on agentic MCP Atlas (83.6%) with 4x faster output. Coding still trails. New Omni video model with conversational editing, physics simulation. Gemini Embedding 2 paper adds multimodal embedding SOTA. Gemini 3 Pro Image (Nano Banana Pro) offers throughput gains. Integrated into Microsoft Copilot with MAI-Code-1. Added native computer use with 4x speed and 80.04% score, plus adversarial training. Gemini 3.5 Live Translate streaming speech-to-speech for 70+ languages. New releases: Nano Banana 2 Lite (fast/cheap image gen, ~3s, $0.03/image) and Gemini Omni Flash (video gen with interactive editing, $0.10/sec). A new Gemini Flash checkpoint appeared on LM Arena (likely 3.6 or 4) — early testers say incremental over 3.5 Flash, no benchmarks yet; Pro delayed. AutomationBench-AA benchmark shows Gemini 3.5 Flash scores on agentic SaaS workflows.