Pistis
A family of multimodal large language models available in 27B and 9B parameter sizes, built on the Qwen3 series. The family uses the IDRL post-training framework and is divided into two specialized variants: Thinking and Agentic.
Timeline (last 90 days)
Pistis is a multimodal large language model family comprising two parameter scales: 27B and 9B.
Pistis's 27B and 9B models are built on Qwen3.6 and Qwen3.5, respectively.
Pistis's training pipeline is divided into two stages: the first stage is large-scale multimodal supervised fine-tuning (SFT), and the second stage is Interleaved Distillation and Reinforcement Learning (IDRL).
Both variants outperform their corresponding base models.
All Facts (4)
Pistis's 27B and 9B models are built on Qwen3.6 and Qwen3.5, respectively.
50%UnverifiedPistis's training pipeline is divided into two stages: the first stage is large-scale multimodal supervised fine-tuning (SFT), and the second stage is Interleaved Distillation and Reinforcement Learning (IDRL).
50%UnverifiedBoth variants outperform their corresponding base models.
50%UnverifiedPistis is a multimodal large language model family comprising two parameter scales: 27B and 9B.
50%