Stepfun introduced the Step 5 model with high performance

Chinese company Stepfun has released a preliminary version of its new large language model Step 5. The developers presented the test results of the novelty within the independent rating Artificial Analysis Intelligence Index version 4.3.

During the evaluation, the model demonstrated a result of 13 points, achieved in 8 hours of computation. The testing was conducted according to a set of metrics, including specialized benchmarks AA-Briefcase, GDPval-AA v2, and SciCode. These indicators allow evaluating the system's ability to solve complex logical and programming tasks in comparison with current market solutions.

A key feature of the release was the positioning of Step 5 on the so-called "Pareto frontier". This means that the model provides a balance between the level of intelligence and the cost of operation. Analysts note that such indicators are critically important for integrating LLM into high-loaded corporate systems, where cost optimization is required for every token.

This release expands Stepfun's presence in the international AI technology market. The model is focused on developers who need tools with predictable performance and high accuracy of responses in technical disciplines. Complete data on the comparison with competitors is available in the current Artificial Analysis report.