The latest AI news and updates of Google Gemma — Google's family of lightweight, open-weight AI models. Covering Google Gemma's latest product updates, company news, and launches from the past 90 days.
GoogleRelated:
Gemini
Google AI Studio
Google GemmaOct 9Google AI Pro and Ultra plans now include premium Colab compute, providing subscribers with A100 80GB or H100 GPUs. This integration supports browser-based execution and fine-tuning of Gemma 4 models, including full fine-tuning for the E4B variant and QLoRA tuning for the 31B model in bf16 precision.
Google GemmaOct 6Google released EmbeddingGemma 2, a 740M-parameter open model that maps text, code, images, video, and audio into a unified embedding space. Built on the Gemma 4 architecture, it features an 8K context window and delivers a 9.92-point improvement on code performance over the original model. The Apache 2.0 model is optimized for on-device RAG and semantic search.
Google GemmaOct 1Google reports that its Gemma 4 31B model achieves GRE learning gains statistically equivalent to expert human tutors. Research involving 2,383 participants found the model matched human performance with a p-value of .015. The AI tutoring cost $0.0052 per percentage point gained, a 918-fold reduction compared to the $4.81 cost of human instruction.
Google GemmaSep 23Google now supports single-command deployment for DiffusionGemma-Jev on Google Cloud Run. This setup provides a Jev API-compatible endpoint with single-step latency of 35-60 ms and batch throughput of 100-123 requests per second. The service costs approximately $3 per hour and scales to zero when idle, removing the need for local GPU hardware during experimentation.
Google GemmaSep 16Google updated the AI Edge Gallery app on macOS to include full support for the Gemma 4 12B model. The update introduces multimodal vision and audio input processing, configurable vision token budgets, and MTP for high-speed inference. These capabilities are optimized to run locally on 16GB Mac hardware, enabling private, offline generative AI workflows.
Google GemmaSep 11Satlyt is running Google’s Gemma 3 1B model directly on satellites to analyze telemetry and diagnose faults locally. This deployment reduces diagnostic data payloads by over 64% by generating compact summaries before downlink. The company plans to launch Gemma 4 E2B on NVIDIA Jetson Orin Nano hardware next to add onboard reasoning and vision capabilities to its orbital infrastructure.
Google GemmaSep 2Google now recommends Gemma 4 as the local model provider for Android Studio. This integration enables offline Agent Mode, ensuring code remains on the local machine without token quotas. The model handles multi-step tasks like feature design, refactoring, and build resolution, with hardware requirements ranging from 8GB to 24GB of RAM depending on the model variant.
Google GemmaSep 1Google reports that its Gemma 4 26B A4B model now runs 130.3% faster on Apple Silicon hardware. This performance gain follows community-driven optimizations on the MLX Fast leaderboard, where developers refined the model's prefill and decode throughput. The update improves local inference efficiency for the 26-billion parameter model on Mac systems.
Google GemmaAug 31Google Gemma is highlighting Gilbeot, an on-device multimodal AI walking assistant for elderly users. The project runs locally on Android phones using the LiteRT-LM framework and custom speech LoRAs to deliver real-time, offline voice navigation. It translates complex map coordinates into natural language instructions based on visible objects, ensuring accessibility on older devices without requiring hardware upgrades.
Google GemmaAug 28Google is running its Gemma 4 E2B model fully locally on an 8GB NVIDIA Jetson Orin Nano. The setup uses native image and audio input alongside function calling to perform vision, voice interaction, and camera control. This demonstration shows multimodal AI capabilities operating entirely within an 8GB memory footprint on power-constrained edge hardware.
Google GemmaAug 22Google reports that its Gemma 4 31B model matches Claude Sonnet 5 on answer quality in AlphaSense benchmarks, while operating at roughly 40x lower cost. This cost-efficiency and low latency enable high-volume AI use cases that are uneconomical with larger frontier models. The benchmark evaluated 245 multi-step finance questions using pre-registered rubrics.
Google GemmaAug 22Google Developer Experts built a real-time AI Race Coach at Sonoma Raceway, running Gemma 4 locally on a Pixel 10. The system integrates Google Antigravity for telemetry orchestration and Gemini for cloud reasoning. This architecture delivers zero-latency audio coaching directly to drivers, processing data at 40 tokens per second to provide split-second track advice during high-speed racing.
Google GemmaAug 12Google demonstrates Gemma 4 E2B running entirely locally on a Raspberry Pi 5 to power the Reachy Mini robot. The system handles voice, vision, and movement offline, achieving text generation speeds of 300 words per minute on 1.5 GB of memory. LiteRT splits workloads between the CPU and GPU to maintain real-time, low-latency performance.
Google GemmaAug 10Google Gemma published the technical report for DiffusionGemma, an experimental open-weight model using discrete diffusion for text generation. By refining blocks of 256 tokens in parallel, it avoids the sequential bottleneck of autoregressive models. It achieves roughly 1,500 output tokens per second on a single NVIDIA H100 GPU while retaining support for multimodal inputs and long contexts.
Google GemmaJul 17Google Gemma released gemma-trainer, a new skill that automates local fine-tuning for Gemma 4 models. The skill enables AI agents to manage training configurations, execute supervised fine-tuning, and evaluate performance. It includes guardrails to prevent capability mismatches and supports exporting models for edge devices using LiteRT-LM.
Google GemmaJul 16Google Gemma makes the Gemma 4 31B model available on LiveKit Inference, optimized for real-time voice agents. The model delivers a 192ms time-to-first-token and 354ms time-to-first-audio. It achieves 76.9% accuracy on the tau2bench agentic tool-use evaluation, surpassing GPT-4.1. This integration provides a low-latency, cost-efficient option for production voice agent workloads.
Frequently asked questions