Gemma 4 pushes AI onto your device

Small models, real power shift

Google says its new open model can beat systems 20x larger.

Gemma 4 is a new family of open models, ranging from 2B to 31B parameters, built for reasoning, agents, and multimodal use across phones and local machines.
The 31B version ranks #3 among open models, and the smaller edge models run offline with 128K context and near-zero latency on devices like phones and Raspberry Pi.

This is about efficiency, not raw scale. The real move is making serious AI run locally, which cuts cost, improves privacy, and gives developers more control without waiting on cloud access.

You can hear the fan spin on a laptop running models that used to need a data center, and that changes who can build and ship AI products quickly.
If this holds, more apps will default to on-device AI, and cloud models start to look like a premium layer rather than the starting point.

If small models keep improving this fast, what still needs the cloud?

Microsoft is quietly reducing reliance

Microsoft just launched three models that overlap with OpenAI.

The company introduced MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, covering speech-to-text, voice generation, and image or video generation.
They are faster in some cases and priced to undercut competitors, with transcription starting at $0.36 per hour and voice at $22 per million characters.

This is a hedge that looks more serious than Microsoft admits.
They are still tied to OpenAI, but building parallel capabilities gives them leverage on pricing, product direction, and long-term control.

You can imagine a product team choosing between two internal options, one from OpenAI and one from Microsoft, both plugged into the same cloud.
If Microsoft keeps improving these models, the partnership shifts from dependence to negotiation.

How long does a partnership last once both sides can replace each other?

OpenAI wants to shape the conversation

OpenAI just bought a daily tech show founders already watch.

The company acquired “TBPN,” a three-hour, five-day-a-week streaming show popular with Silicon Valley executives, and plans to keep it running under its strategy team.
The move comes as criticism of AI grows, and OpenAI admits its usual communications approach is not working.

This is a distribution play. If you cannot rely on press coverage or public trust, owning a channel with built-in audience and friendly framing starts to look practical.

You can picture a studio with bright lights and founders rotating through, except now the platform sits inside the company they may need to question.
Even with promises of independence, access and tone tend to shift once ownership changes, and rival CEOs may think twice before showing up.

If AI companies start owning media, who is left to question them from the outside?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.