Google has officially introduced Gemma 3, the latest evolution in its Gemma family of open-weight AI models, succeeding last year’s Gemma 2. This new release boasts enhanced efficiency, strong benchmark performance, and expanded capabilities across multiple modalities.

Gemma 3 is available in four parameter sizes—1B, 4B, 12B, and 27B—offering flexibility for different use cases. It supports an extended context window of 128K tokens, allowing for better processing of long-form data. The model is multimodal, capable of analysing text, images, and videos, and supports 35 languages natively while offering pre-trained support for 140 languages.

In benchmark evaluations, Gemma 3 27B excelled in the Chatbot Arena, outperforming models such as DeepSeek-V3, OpenAI’s o3-mini, and Meta’s Llama 3-405B through human-judged, side-by-side comparisons. It also delivered competitive scores across industry-standard tests:

• MMLU-Pro: 67.5%
• GPQA Diamond: 42.4%

For comparison, Claude 3.5 Haiku scored 63% on MMLU-Pro and 41% on GPQA Diamond, while GPT-4o Mini achieved 65% and 43%, respectively. However, Meta’s Llama 3.3 70B led the rankings with 71% on MMLU-Pro and 50% on GPQA Diamond.

A standout feature of Gemma 3 is its exceptional compute efficiency. Google highlighted that Gemma 3 27B achieved its benchmark results using a single NVIDIA H100 GPU, whereas competing models required up to 32 GPUs. This efficiency stems from architectural refinements, including KV-cache memory optimizations, which enhance performance while maintaining lower computational demands.

Also Read :- Lip-Bu Tan appointed as Intel’s new CEO

Google has released a detailed technical report outlining the model’s architecture, optimizations, and benchmarks. Users can access Gemma 3 via multiple platforms, including:

• Google AI Studio (web & API access)
• Google GenAI SDK
• Hugging Face, Ollama, and Kaggle (for local deployment)

Alongside Gemma 3, Google has launched ShieldGemma 2, a 4B parameter AI model designed to detect and flag harmful images. Built on Gemma 3’s foundation, ShieldGemma 2 provides safety labels for potentially dangerous, explicit, or violent content, reinforcing ethical AI deployment.

With Gemma 3, Google is setting new benchmarks for inefficient AI development while broadening its capabilities across text, image, and video analysis—a move that could reshape the future of AI applications.

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!