Google's Gemma 4 Goes Open-Source with Major Parameter Expansion

As Chinese tech firms continue to hold a strong position in the global open-source large model arena, US tech giants are seeking to reclaim influence through differentiated strategies.
Recent media reports indicate that Demis Hassabis, CEO of Google DeepMind, recently hinted on a social platform with a "four diamonds" icon that the new open-source model Gemma 4 is nearing its official launch. This comes exactly one year after its predecessor, Gemma 3, aligning with Google's established release cadence in the large model space.
Major Specs Upgrade: New 120B Model Pushes Local Deployment Limits
Compared to the previous generation, Gemma 4 represents a significant leap in scale:
Quadrupled Parameters: Rumors suggest this version will introduce a large model with 120 billion parameters, four times the size of the last generation.
MoE Architecture: To balance performance and efficiency, the model is expected to utilize a Mixture of Experts (MoE) architecture, with only 15 billion active parameters. This design could allow even the large-parameter model to run locally on consumer-grade GPUs.
Capability Evolution: Gemma 4 is anticipated to double its context window and demonstrate significantly deeper logical reasoning and complex task execution abilities.
Strategic Play: Containing "Chinese Influence" in the Open-Source Community
FastTech analysis notes that while the current focus for US giants is on closed-source business models, Google is strategically releasing technical advancements to prevent Chinese companies from dominating the open-source ecosystem entirely:
Timing Strategy: Google's decision to release the open-source version roughly six months after its flagship closed-source Gemini 3.0 series allows it to protect commercial revenue while maintaining developer community influence through open-source projects.
The Localization Moat: The core positioning of Gemma 4 remains "local deployment." By optimizing lightweight model performance, Google aims to compete directly with domestic open-source models on end-user experience without compromising its core commercial interests.
Industry Outlook: The Open-Source Race Enters an Era of "Parameters and Efficiency"
With the arrival of Gemma 4, the competitive bar for open-source large models has been raised. Industry observers widely agree that while open-source may not be Google's top priority, its profound algorithmic expertise remains a formidable variable. Whether Gemma 4 can outperform leading domestic open-source models at a comparable parameter scale will be a key focus for the global AI community in the second half of the year.
Related article
Siri AI Settlement Lets US iPhone Users Install Third-Party Apps, Receive $25 Per Device
Following a class-action lawsuit in the U.S. over delayed Siri AI integration, Apple has launched an official settlement portal for eligible iPhone owners to file compensation claims.In May, Apple agreed to a $250 million (approx. RMB 1.6 billion) se
Anthropic launches Opus 5
Anthropic unveiled Opus 5 on Friday, the latest iteration of its flagship model. Although smaller than Fable 5, this version offers lower costs and fewer restrictions, making it the preferred choice for most applications.Notably, Opus 5 surpasses Fab
AI Hallucination Nearly Triggers US Military Operation
During the spring, U.S. officials made a startling discovery while military aircraft were already airborne: the intelligence guiding an armed strike against a Chinese vessel was entirely fabricated by an AI chatbot. The operation was called off at th
Related Special Topic Recommendations
Comments (0)
0/500

As Chinese tech firms continue to hold a strong position in the global open-source large model arena, US tech giants are seeking to reclaim influence through differentiated strategies.
Recent media reports indicate that Demis Hassabis, CEO of Google DeepMind, recently hinted on a social platform with a "four diamonds" icon that the new open-source model Gemma 4 is nearing its official launch. This comes exactly one year after its predecessor, Gemma 3, aligning with Google's established release cadence in the large model space.
Major Specs Upgrade: New 120B Model Pushes Local Deployment Limits
Compared to the previous generation, Gemma 4 represents a significant leap in scale:
Quadrupled Parameters: Rumors suggest this version will introduce a large model with 120 billion parameters, four times the size of the last generation.
MoE Architecture: To balance performance and efficiency, the model is expected to utilize a Mixture of Experts (MoE) architecture, with only 15 billion active parameters. This design could allow even the large-parameter model to run locally on consumer-grade GPUs.
Capability Evolution: Gemma 4 is anticipated to double its context window and demonstrate significantly deeper logical reasoning and complex task execution abilities.
Strategic Play: Containing "Chinese Influence" in the Open-Source Community
Timing Strategy: Google's decision to release the open-source version roughly six months after its flagship closed-source Gemini 3.0 series allows it to protect commercial revenue while maintaining developer community influence through open-source projects.
The Localization Moat: The core positioning of Gemma 4 remains "local deployment." By optimizing lightweight model performance, Google aims to compete directly with domestic open-source models on end-user experience without compromising its core commercial interests.
Industry Outlook: The Open-Source Race Enters an Era of "Parameters and Efficiency"
With the arrival of Gemma 4, the competitive bar for open-source large models has been raised. Industry observers widely agree that while open-source may not be Google's top priority, its profound algorithmic expertise remains a formidable variable. Whether Gemma 4 can outperform leading domestic open-source models at a comparable parameter scale will be a key focus for the global AI community in the second half of the year.
Siri AI Settlement Lets US iPhone Users Install Third-Party Apps, Receive $25 Per Device
Following a class-action lawsuit in the U.S. over delayed Siri AI integration, Apple has launched an official settlement portal for eligible iPhone owners to file compensation claims.In May, Apple agreed to a $250 million (approx. RMB 1.6 billion) se
Anthropic launches Opus 5
Anthropic unveiled Opus 5 on Friday, the latest iteration of its flagship model. Although smaller than Fable 5, this version offers lower costs and fewer restrictions, making it the preferred choice for most applications.Notably, Opus 5 surpasses Fab
AI Hallucination Nearly Triggers US Military Operation
During the spring, U.S. officials made a startling discovery while military aircraft were already airborne: the intelligence guiding an armed strike against a Chinese vessel was entirely fabricated by an AI chatbot. The operation was called off at th





Home






