Google has launched Gemma 3, the third generation of its open-source AI models. The model is better than rivals like DeepSeek-V3, o3-mini of OpenAI, and Meta's Llama 3-405B. The new Gemma 3 was launched on March 13, 2025. It builds on the success of Gemma 2 while improving efficiency, benchmark scores, and compute optimization.
Gemma 3 is available in several parameter sizes, such as 1B, 4B, 12B, and 27B, and offers a 128K-token context window to enable improved long-form reasoning. Gemma 3 is also a multimodal model that has the ability to process video, images, and text and supports pre-trained multilingual capabilities with 140 languages, including support for 35 languages out-of-the-box.
In standard AI benchmark tests, Gemma 3 27B recorded a 67.5% score on MMLU-Pro and 42.4% on GPQA Diamond, surpassing OpenAI’s o3-mini and Claude 3.5 Haiku. For comparison, Claude 3.5 Haiku achieved 63% on MMLU-Pro and 41% on GPQA Diamond, while OpenAI’s GPT-4o Mini scored 65% and 43% on the same benchmarks. However, Meta’s Llama 3-70B emerged as the highest scorer, with 71% on MMLU-Pro and 50% on GPQA Diamond.
The biggest advantage of Gemma 3 is its higher efficiency. While all other top AI models need enormous computing power, Gemma 3 27B does the same benchmark results with only one NVIDIA H100 GPU, a world of difference from rivals that use up to 32 GPUs to do the same amount of work.
For optimal efficiency, Google has re-engineered the architecture of the model so that KV-cache memory increase is reduced to a minimum, providing greater memory efficiency in dealing with long contexts. This makes Gemma 3 one of the most efficient models on offer for developers and researchers.
Google has highly available Gemma 3 on all platforms:
Google AI Studio (web-based access)
Google GenAI SDK
API access for developers
Downloadable local deployments through Hugging Face, Ollama, and Kaggle.
Google has also introduced ShieldGemma 2, a 4B parameter AI model tailor-made for image safety detection. Created from Gemma 3, ShieldGemma 2 assists in identifying unsafe, explicit, or violent images with the use of AI-based content safety labeling.
Google has created a new standard with Gemma 3 for AI efficiency and performance. With top-of-the-line results being delivered with low compute power, the model is set to see widespread adoption in research and enterprise use.
As the war between Google, OpenAI, and Meta intensifies, the AI models are getting more advanced. Each is pushing the performance, scalability, and cost-effectiveness boundaries in AI.