SubQ Snap 1.2 tops Artificial Analysis index

Specialized decision models are exposing the inefficiency of routing routine agentic judgments to full-scale generative LLMs. Subquadratic's linear attention and order-robust scoring provide a fast, budget-friendly primitive for high-throughput evaluation and routing pipelines.

• –Subquadratic Sparse Attention avoids quadratic compute bottlenecks, crunching 100K tokens in under a second
• –Inference-time option-order invariance resolves the classic transformer vulnerability to answer option shuffling
• –Priced at $0.04 per million input tokens, it slashes costs compared to general-purpose frontier models for triage tasks
• –A 384K multimodal context window accommodates entire codebases and document troves without chunking