Pocket Power : From State of the Art to Your Phone in 23 Months | Tomasz Tunguz

Source: original

Pocket Power : From State of the Art to Your Phone in 23 Months

April 5, 2026

AI data

Two years ago, the idea of useful AI on your phone was fantastical. Siri couldn’t finish a sentence. Local models hallucinated nonsense.

Last week, Google released Gemma 4 E4B1, a free model that matches GPT-4o and runs entirely on your phone.2

The next few weeks promise even more advanced pocket models. The market expects new releases from DeepSeek3, Qwen4, Kimi5 & Minimax6.

Frontier models don’t stay frontier for long. Within three to four months, you can run a model with similar performance on your laptop; 23 months later, you can run the same model on your phone.

Parameters Required for GPT-4o-Level HumanEval Score : 450x compression in 23 months

Three forces are driving this compression. Better algorithms : distillation & reinforcement learning squeeze more capability into fewer parameters. Talent density : the biggest prizes in capitalism attract the best minds in the field. These are the fastest growing software companies in history. And capital : a trillion dollars invested in data centers powering training.

In 23 months, the same capability that needed 1.8 trillion parameters now fits in 4 billion parameters. A 450x compression. At this rate, the phone in your pocket will run today’s frontier models before you upgrade it. 1. Google AI Edge Gallery on iOS App Store ↩︎

  1. Gemma 4 E4B matches or exceeds GPT-4o across multiple benchmarks including MATH, GSM8K, GPQA Diamond & HumanEval. Full benchmark comparison ↩︎

  2. DeepSeek’s new AI model ↩︎

  3. Qwen 3.6 ↩︎

  4. Kimi K3 ↩︎

  5. MiniMax M2.5 release ↩︎