Models & Research
Top in Models & Research
Google launches Gemini 3.8 Live and 3.8 Live Extended Thinking models
3 min read16 Sept 2026
New model releases, benchmarks and research from the frontier labs and the open-source community.
Top in Models & Research
ByteDance Seed's new HarnessDev benchmark finds AI models can build agent tool-chains that write well but trail human-built references on coding and search, and that only 34 of 64 self-improvements generalised.
Cohere has released North Small Translate, an open-weight 218-billion-parameter translation model that the company says outperforms DeepL and Google Translate on its own WMT26 evaluation. Commercial licensing runs through RWS.