Google: Gemma 4 26B A4B

Google's Gemma 4 26B A4B IT, an instruction-tuned Mixture-of-Experts (MoE) model, demonstrates strong performance across key metrics. It consistently ranks among the fastest models, placing in the 82nd percentile for speed, and offers competitive pricing, landing in the 78th percentile for cost-effectiveness. A standout feature is its exceptional reliability, achieving a 100% success rate across benchmarks, indicating minimal technical failures and consistent response delivery. In terms of specific capabilities, the model exhibits a high degree of accuracy in acknowledging uncertainty, scoring 98.0% on the Hallucinations (Baseline) benchmark. This indicates a strong ability to identify and appropriately respond to fictional concepts by selecting "I don't know," a crucial aspect for trustworthy AI. This performance is achieved at a cost of $0.0022 and a duration of 76933ms, both of which are competitive within their respective percentiles. Its architectural design, activating only 3.8B parameters out of 25.2B total, allows it to deliver near-31B quality at a significantly reduced computational cost. The model's support for multimodal input, a 256K token context window, native function calling, and configurable reasoning further enhance its versatility and utility.