OpenAI: gpt-oss-20b

The OpenAI gpt-oss-20b model, an open-weight 21B parameter Mixture-of-Experts (MoE) model, demonstrates exceptional speed and competitive pricing. It consistently ranks among the fastest models, achieving an Infinityth percentile across 17 benchmarks, and offers among the most competitive pricing, also at an Infinityth percentile across 9 benchmarks. This optimization for lower-latency inference and deployability on consumer or single-GPU hardware is evident in its performance. In terms of specific benchmarks, gpt-oss-20b shows strong performance in several areas. It achieves high accuracy in Ethics (99.0%), General Knowledge (99.0%), and Coding (92.0%), indicating robust understanding in these domains. Its Instruction Following capabilities are also solid at 66.0% accuracy, and Reasoning at 89.8%. Notably, it achieved perfect accuracy in two instances of Keyword Topic Relevance Classification, with one being the fastest recorded duration. However, the model exhibits significant weaknesses in other Keyword Topic Relevance Classification benchmarks, scoring 0.0% accuracy in multiple instances, suggesting inconsistency or sensitivity to specific test variations. Its Mathematics performance is moderate at 66.7% accuracy, but with a very high duration. The model also shows a tendency to hallucinate, with 70.0% accuracy in the Hallucinations benchmark, indicating it doesn't always correctly identify fictional concepts.