The whole dayWed 28 Oct 2026
AI Compass

Particle launches Particle-1 token-efficient software engineering model

Released on 28 October 2026, Particle-1 is post-trained on DeepSeek V4.1 Flash to cut output tokens by roughly 45% across software engineering benchmarks without dropping pass@1 accuracy.

Logo: ParticleCover by AI Compass

Particle released Particle-1 on 28 October 2026, a software engineering model post-trained on DeepSeek V4.1 Flash using Cascade RL and GSPO with a rank-64 LoRA adapter. Across DeepSWE 1.1, Terminal-Bench 2.1, ReactBench, and ADE Bench, Particle-1 matches base pass@1 accuracy while reducing output tokens by 28.4% to 52.2% and cutting task cost by up to 51.6%, priced at $0.25 per million input tokens and $1 per million output tokens on the Particle API and OpenRouter.

Sources

  1. PrimaryParticle blogparticle.ai
  2. OfficialParticle model documentationdocs.particle.ai

The rest of the day

Read the whole day

01ModelGoogle DeepMind

Google DeepMind unveils Gemini 4 Carbon frontier model

Google DeepMind’s next frontier model after Gemini 4 Argon reaches #1 on the Artificial Analysis Intelligence Index v5.0 and begins rolling out to trusted cyber defenders through the Fairwind Program ahead of general availability.

Spotted an error? Email aiforsuccess.in+compass@gmail.com or tag @ai_for_success on X