StorySoftware
Google releases Gemma 4, open-weight models from 2B to 31B, with Apache 2.0 license
Google released Gemma 4, its latest family of open-weight language models, with four sizes from 2.3B to 31B parameters under the Apache 2.0 license. The 31B model scores 89% on AIME 2026 and 80% on LiveCodeBench, placing it third globally on the Arena AI leaderboard. All sizes support multimodal inputs including text, images, audio and video.
- Four model sizes available: 2.3B, 4.5B, 26B mixture-of-experts, and 31B dense.
- 31B model scores 89.2% on AIME 2026 math benchmark, up from Gemma 3's 20.8%.
- Multimodal: all sizes accept text, images, audio; larger models add video support.
- Apache 2.0 license permits commercial use without restrictions.
- Includes function calling and structured JSON output for building autonomous agents.