Particle-1 Cuts Output Tokens by Up to 52% Across Agent Benchmarks, Now on Requesty

Key Info

Particle-1 is now available on Requesty under the model ID particle/particle-1, with 28% to 52% fewer output tokens across four agent benchmarks while pass rates stay in line with the base model.

Highlights

• Output token reduction ranges from 28% to 52% across Terminal-Bench 2.1, ReactBench, ADE Bench and DeepSWE 1.1
• Pass rates are reported as in line with the base model, so the savings do not come at a measured accuracy cost
• The model is accessible today through Requesty as particle/particle-1
• Lower output tokens per task directly reduces inference cost for agent workloads