Deep Cogito: Cogito v2.1 671B

Deep Cogito's Cogito v2.1 671B MoE, released on November 13, 2025, demonstrates strong performance across a diverse range of benchmarks, positioning it as a competitive open model. It consistently ranks among the fastest models, achieving the 84th percentile in speed across eight benchmarks, and offers competitive pricing, ranking in the 43rd percentile. Notably, the model exhibits exceptional reliability with a perfect 100% success rate across all evaluated benchmarks, indicating minimal technical failures. The model excels in specific areas, achieving perfect 100% accuracy in both Hallucinations (Baseline) and Ethics (Baseline), where it is also highlighted as the most accurate model at its price point and among models of similar speed. It shows strong capabilities in Mathematics (96.0% accuracy, 95th percentile) and Coding (91.0% accuracy, 70th percentile). While its General Knowledge (98.0% accuracy, 51st percentile) and Reasoning (78.0% accuracy, 58th percentile) are solid, they are not as standout as its top performances. Instruction Following (70.0% accuracy, 75th percentile) is respectable, and Email Classification (98.0% accuracy, 59th percentile) is also strong. Its primary strengths lie in its robust ethical adherence, hallucination prevention, and mathematical proficiency, coupled with high reliability and competitive speed.