Research
Benchmark Study on Instruction Following in LRMs
Together AI Bloghigh confidence
Why it matters
- →This benchmark provides a systematic way to evaluate LRM performance in following instructions. • Understanding instruction adherence can lead to improvements in model design and user interaction. • The findings may influence future research and development in large reasoning models.
Together AI has introduced ReasonIF, a benchmark designed to assess how well large reasoning models (LRMs) follow user instructions throughout their reasoning processes. The study found that models like GPT-OSS-120B and Qwen3-235B fail to adhere to instructions more than 75% of the time, with performance degrading as task difficulty increases. ReasonIF consists of 300 math and science problems paired with specific reasoning instructions, aiming to improve controllability and transparency in model outputs. This research highlights the need for better instruction adherence in LRMs to enhance their usability and reliability.
Read originalMore from Together AI Blog
.png)
© Together AI BlogModels & Labsother
Together AI Partners with Adaption
Together AI and Adaption have formed a partnership to integrate Together Fine-Tuning into Adaptive Data, enabling teams to optimize datasets and deploy stronger open models.
Together AI Blog
© Together AI BlogGeneral AIother
