This is a news item from our media watch, potentially of interest to part of the university community. The content relayed does not reflect a position of UNIL.
In April 2026, three laboratories released freely accessible and particularly high-performing language models: Google’s Gemma 4, Moonshot AI’s Kimi K2.6 and DeepSeek V4.
- https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/
- https://www.kimi.com/blog/kimi-k2-6
- https://api-docs.deepseek.com/news/news260424
According to the independent Artificial Analysis ranking (https://artificialanalysis.ai/models), Kimi K2.6 achieves an intelligence score of 54, compared to 60 for OpenAI’s GPT‑5.5. The gap with commercial models continues to narrow.
Unlike the latter, open models can be deployed on institutional infrastructure, which raises new questions regarding data privacy and digital sovereignty.