Krux
April 4, 2026
Google's Gemma 4 Brings 256K Context to Open Models
Published: April 4, 2026 at 12:37 AM
Updated: April 4, 2026 at 12:37 AM
100-word summary
Google just released Gemma 4, a four-model family under the Apache 2.0 license. The headline feature: context windows up to 256K tokens, letting open models chew through entire codebases or legal documents without proprietary API bills. The smallest model handles 128K contexts on edge hardware. It's available now on Hugging Face, Kaggle, and Ollama with native function-calling for building agents that interact with tools. The 26B mixture-of-experts model activates only 3.8B parameters per query, keeping responses fast. What's striking is the pairing: truly open licensing with capabilities previously locked behind commercial models. Open-source AI just got a lot more competitive with closed alternatives.
What happened
Google just released Gemma 4, a four-model family under the Apache 2.0 license. The headline feature: context windows up to 256K tokens, letting open models chew through entire codebases or legal documents without proprietary API bills. The smallest model handles 128K contexts on edge hardware. It's available now on Hugging Face, Kaggle, and Ollama with native function-calling for building agents that interact with tools. The 26B mixture-of-experts model activates only 3.8B parameters per query, keeping responses fast. What's striking is the pairing: truly open licensing with capabilities previously locked behind commercial models.
Why it matters
Open-source AI just got a lot more competitive with closed alternatives.