InternLM

A multimodal model aimed at scientific applications.

A 1T model by InternLM/Shanghai AI Laboratory, which focuses on STEM. However, its real utility seems to be below what the benchmarks suggest.

One of the first multimodal Qwen3 fine-tunes. This release from the talented InternLM team combines the large Qwen3 MoE with their own ViT.

A bit of a different take on the current rage of reasoning models. Quoting the model card: "Our method leverages best-of-N (BoN) sampling for behavior cloning and reshapes negative sample rewards to ensure gradient consistency. Also, to address the challenge of sparse rewards in long chain-of-thought reasoning, we incorporate an on-policy token-level reward model that identifies key tokens in reasoning trajectories for importance sampling." There are more details in the paper, but mostly this goes to show that reward models (specifically, outcome reward models) are still important to reasoning and it isn't that "explicit verification is all you need."

A bit of a different take on the current rage of reasoning models. Quoting the model card: "Our method leverages best-of-N (BoN) sampling for behavior cloning and reshapes negative sample rewards to ensure gradient consistency. Also, to address the challenge of sparse rewards in long chain-of-thought reasoning, we incorporate an on-policy token-level reward model that identifies key tokens in reasoning trajectories for importance sampling." There are more details in the paper, but mostly this goes to show that reward models (specifically, outcome reward models) are still important to reasoning and it isn't that "explicit verification is all you need."

A multi-modal reward model.

Similar to the other Chinese labs, InternLM has also updated their model series before Chinese New Year.

Another very strong reward model on RewardBench. Trained on 3.5 million preference pairs, which is way more than most. A good trend.

Next model in the popular series of math models.