Inference.net: Schematron V2 Turbo

Inference.net's Schematron V2 Turbo, a 3B-parameter HTML-to-JSON extraction model, demonstrates exceptional operational efficiency. It consistently ranks among the fastest models and offers highly competitive pricing across all benchmarks, making it a strong contender for high-volume extraction workloads. The model also exhibits outstanding reliability with a 99% success rate, indicating minimal technical failures. However, its performance on general cognitive tasks is notably low. It struggles significantly with Instruction Following (0.0% accuracy) and Reasoning (14.0% accuracy). While its accuracy in Hallucinations (58.0%), General Knowledge (43.5%), Ethics (66.0%), Mathematics (49.0%), Email Classification (58.0%), and Coding (49.0%) is generally below average, its primary design as an HTML-to-JSON extraction model suggests these benchmarks may not reflect its core strength. Its key strength lies in its speed, cost-effectiveness, and reliability, which are crucial for its intended high-throughput extraction applications.