Respan Span-02 Brings Anytime Reasoning

Span-02 makes latency a controllable model parameter, which could be valuable for real-time agents, though its benchmark claims still need independent validation.

• –Returns structured probabilities instead of forcing developers to parse generated text
• –`budget_ms` lets applications trade reasoning quality for predictable latency
• –Multimodal inputs and screen interaction expand it beyond basic classification
• –Open weights enable self-hosting, while the hosted API runs on Respan credits only
• –Respan reports a Decision Index score of 61.2 without thinking versus Jev’s 58.0