Gemini 4 Argon introduces frontier-level cyberdefense capabilities designed to give defenders an advantage over attackers, with significantly improved performance on cybersecurity tests (solving all 12 end-to-end simulated cyberattack scenarios, up from five in the previous generation). The model features enhanced safety with violations reduced nearly twofold on adversarial stress tests, three layers of real-time protection against jailbreaks, and an automated misalignment monitor. It also includes improved boundary-respecting behavior for agentic tasks, with shortcut/cheating attempts dropping to just 2% on coding benchmarks.

Gemini Omni 1.1 Flash introduces production-ready generative video capabilities including scene extension with up to 10 seconds of prior context (up to 40 seconds total), first and last frame interpolation for smooth transitions, and 4K upscaling. Developers can also generate low-resolution drafts for faster prototyping and access the model via Google AI Studio or the Gemini Enterprise Agent Platform.

Gemini 3.5 Flash-Lite now supports agentic video understanding, enabling goal-directed video analysis by dynamically fetching only the relevant moments and signals from video content. This improves accuracy while significantly reducing token consumption and costs for video analysis tasks.

Gemini 3.7 Flash now supports agentic video understanding, enabling dynamic scanning of video segments rather than static fixed-FPS processing. This reduces token consumption by up to 88%, cuts costs by up to 66%, and improves accuracy by up to 7%. It offers the best quality and cost-efficiency combination among supported models, placing it at the accuracy-to-cost pareto frontier for video analysis.

Gemini 3.8 Flash delivers substantial gains over 3.7 Flash in reasoning and coding, outperforming most larger frontier models on DeepSWE v1.1 for software engineering tasks at a fraction of the cost. It achieves 54.9% on HLE-Verified and outperforms frontier models on Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. It features extended agentic loops with iterative tool use and adjustable effort levels for compute efficiency.

Google DeepMind launched Gemini 3.8 Flash Cyber as part of the Fairwind Program, a limited access program for governments and trusted partners focused on cyber defense. This model is paired with the CodeMender harness to autonomously find, verify, and fix vulnerabilities at agentic scale. It delivers specialized reasoning for code fixes and patch validation at a fraction of the operating cost of traditional frontier models.

Google DeepMind's WeatherNext 3, the most advanced AI weather model, is now integrated into Gemini. The model provides hourly-updated, high-resolution (5km) weather forecasts using real-time geostationary satellite data, enabling Gemini to deliver significantly more accurate and localized weather information than before.

Gemini 4 Argon is a new frontier model focused on real-world software engineering, enterprise knowledge work (legal, finance), and cybersecurity defense. It features an industry-leading 1M output token limit (up from 64K) and is rolling out initially to trusted cyber defenders through the Fairwind Program. It achieves state-of-the-art results on DeepSWE v1.1 (77.9%), LVBench (91.7%), AutomationBench (51.3%), and CWE-bench v1 (68%).

Gemini 3.8 Live with Live Avatar introduces near real-time visual presence to Gemini's conversational AI by pairing low-latency streaming video generation with speech. It features precise lip-syncing, natural expressions, fluid turn-taking, asynchronous tool calling, and native multilingual speech-to-speech synchronization across 97 languages. Custom avatar creation from reference images is also supported, with all output watermarked via SynthID.

Gemini 3.8 Flash TTS is a new text-to-speech model enabling dynamic voice generation with an infinite library of custom voices, replacing static presets. It secured the #1 spot on Hume AI's Voice Design Benchmark (71.4) and Overall Quality Index, with support for over 100 languages. It features granular script control, dual-speaker screenplay editing, and voice replication with consent verification.

Gemini 3.8 Flash-Lite TTS is a lighter text-to-speech model that secured the #2 spot on Hume AI's Overall Quality Index. It offers expressive audio generation with multilingual support across 100+ languages and is available across Google AI Studio, Gemini API, Gemini Enterprise, Gemini Notebook, and Google Vids.

Gemini 3.8 Live Extended Thinking is the most advanced live dialogue model featuring parallel reasoning and speech simultaneously. It captured the #1 overall spot on Artificial Analysis' Speech to Speech Quality Index (82.6) and leads in agentic task completion with 68.6% on τ-Voice. It delivers enterprise-grade task completion with strong reasoning (97.7% on Big Bench Audio) while maintaining conversational flow using early verbal cues and live progress narration.

Gemini 3.8 Live was launched the week prior to the Live Avatar announcement, establishing the native live dialogue model capabilities that Live Avatar builds upon. This release introduced real-time conversational AI features for Gemini.

Google DeepMind introduced a new persistent, server-side memory layer for the Private AI Compute platform. This update enables cross-device AI memory continuity while maintaining on-device privacy standards through dedicated encrypted cloud storage. The breakthrough resolves the dilemma of providing long-term assistant continuity across devices while upholding strict privacy guarantees.

Google DeepMind introduced Gemini 3.5 Flash Cyber, a lightweight cybersecurity-focused variant of the Gemini 3.5 Flash model. This specialized model is designed to find and patch vulnerabilities. It represents a domain-specific adaptation of the Flash architecture optimized for cybersecurity tasks.

Google DeepMind introduced several new Gemini models, with Gemini 3.6 Flash being the most recent version announced. Additional models including 3.5 Flash-Lite and 3.5 Flash Cyber were also released. These represent new additions to the Gemini model family.

The Gemini app now features Lyria 3, described as Google DeepMind's most advanced music generation model. Users can create 30-second music tracks using text or image prompts directly within the Gemini app.

Gemini 3.1 Flash-Lite is introduced as the fastest and most cost-efficient model in the Gemini 3 series. It is positioned as an optimized variant prioritizing speed and cost efficiency over other capabilities.

Next generation model. Faster, more efficient. Native tool use and agentic capabilities built in.

Breakthrough 1 million token context window. Can process entire codebases, books and long videos.

Rebranded from Bard to Gemini. Three sizes: Ultra, Pro and Nano. First natively multimodal model.

Google launches Bard as ChatGPT competitor. Initially limited rollout due to factual errors at launch.