UkisAI releases two Swift 1.5 reasoning-efficiency models
UkisAI announced Swift 1.5 GLM-5.3-Flash and Swift 1.5 Ornith-9B, with GGUF and other quantized weights available for local inference. The company says GLM-5.3-Flash cut thinking-token usage by 61.4% on MMLU-PRO with a 0.46% average accuracy loss, while Ornith-9B reduced thinking-token usage by 33.5% without an accuracy loss. These results are vendor claims based on UkisAI's own benchmarks and have not been independently verified.
The release provides locally runnable reasoning models optimized for fewer thinking tokens, a capability relevant to inference cost and latency, although the reported efficiency and accuracy results still require independent validation.
#UkisAI#Swift 1.5 GLM-5.3-Flash#Swift 1.5 Ornith-9B#GLM-5.3-Flash#Ornith 9B#GSQ-RCO#MTP#DFLASH2#MMLU-Pro