South Korea’s artificial intelligence (AI) startup Upstage announced on Feb. 22 that it released the beta version of its open-source small language model (SLM), Specialized and Optimized LLM and Applications with Reliability (Solar).
Through simple application programming interface (API) calls, users can access both general conversational models and translation-focused models. These services will be available for free until Mar. 31.
Solar is equipped with a speed more than 2.5 times faster than OpenAI's GPT-3.5.
It can handle up to 4,000 tokens per prompt, with plans to expand this capacity to a staggering 64,000 tokens. T
Tokens represent the smallest unit of meaningful data in a dataset.
Upstage's translation-optimized model shines in Meta's "FLORES" benchmark test, surpassing even established models like GPT-4 and DeepL.
This impressive accuracy stems from Solar's ability to understand context and infer comprehensive meaning, going beyond simple sentence-by-sentence translation.