Particle-1 launches token-efficient coding on AI Gateway

Particle-1 makes token efficiency a first-class model capability, which could matter more than marginal benchmark gains for production coding agents.

• –Particle claims 28.4%–52.2% fewer output tokens and 34.4%–51.6% lower task costs than DeepSeek V4.1 Flash
• –Reinforcement learning encourages shorter reasoning, fewer repair loops, and faster task completion
• –Cloudflare provides a 1M-token context window, image input, streaming, tool use, and configurable reasoning effort
• –The initial focus is software engineering and data engineering, with broader domain coverage still unproven
• –Results are vendor-reported, so independent evaluations will determine whether the efficiency gains generalize