Swedish Tech Scene Benefits from Breakthrough AI Model Cutting Costs by 50%

Swedish AI technology has taken a major leap forward with the introduction of DeepSeek's experimental language model, DeepSeek-V3.2-Exp, which significantly reduces the cost and computational demand of processing long text sequences. AI models like ChatGPT traditionally slow down during lengthy discussions due to the exponential increase in operations required when relating each word to every other word — for example, a 1,000-word query demands a million operations, escalating to 100 million operations for 10,000 words.

DeepSeek’s innovation, called DeepSeek Sparse Attention (DSA), addresses this challenge by focusing computation only on a carefully selected subset of relevant word relationships rather than all possible pairs. This enables the model to analyze complex long texts more efficiently. According to a report in Dagens PS, this approach has cut API usage costs by 50% in long-context tasks, a groundbreaking reduction in expense for AI services. Notably, DeepSeek’s sparse attention achieves a fine-grained analysis thanks to a 'lightning indexer' ranking word pair relevancy — a first of its kind.

The DeepSeek-V3.2-Exp model has been made available as open source under the MIT license, encouraging further research and development. This development complements recent Swedish AI industry momentum, exemplified by the recent sale of Sweden’s Sana to Workday for billions, underscoring growing global investment in Swedish AI capabilities.

This breakthrough stands to impact wide-ranging applications reliant on efficient AI language processing, potentially benefiting developers, businesses, and consumers across Sweden and beyond.

According to Dagens PS, "The company has cut API costs by 50% due to improved efficiency from the new model," marking a significant milestone in AI cost-efficiency technology.