Definition
System One models perform fast structured decisions — parallel to Kahneman System 1 intuition — vs System Two reasoning LLMs. Optimized for classification, routing, safety gates, and bounded computer-use actions.
Key Points
- 2026-10-07: AWS strands-decider-2b — 2B open-source decision model with pointer head on Qwen3.5-2B; ~115ms latency; Apache-2.0 (decision-models)
- 2026-09-18/19: cua CUA-S1-FORMS — 706K params, scores predefined form actions (value/check/click/skip) in one forward pass; inspired by jev (2026-09-19-cua-s1-computer-use-model)
- TypeSafe jev pioneered calibrated System One decision models for enterprise workflows (2026-09-18-typesafe-jev-calibrated-decision-model)
Related
- cua
- computer-use
- jev
- typesafe-ai
- rlcd
- calibrated-decisions
- llm-reliability
- strands-decider-2b
- decision-models
- agent-routing