ModelInternVL3.5
InternVL3.5-8B
An 8-billion-parameter vision-language model in the InternVL series, designed to support image understanding and multimodal reasoning tasks.
Timeline (last 90 days)
Oct 9
On the Qwen3-VL series of models (from 2B to 30B-A3B) and on InternVL3.5-8B, SPLIT-RL improves average accuracy by 1.4 to 6.1 percentage points compared to standard GRPO.
Unverified50%