More System 1 model releases

Jacky Kwok introduces CLM, trained with a contrastive objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and claims up to 9× faster inference than Jev. The tweet links to a blog post on the architecture.

Fastino Labs releases GLiNER2.5-Decide, a 340M-parameter open-weight encoder model for fast, deterministic classification.

Laya is an open-source, Jev-like decision model whose author says they built it with RL a year earlier, in a post titled “I Built Non-Autoregressive Decision Models with RL a Year Ago. Then a Frontier Lab Called It a ‘Breakthrough’.” People are already fine-tuning and porting it, such as this MLX port that plays Snake at 60 decisions per second on an M3 Max.