Open RL fine-tune of Qwen3-VL 8B Inst. Matches proprietary models on 30 benchmarks. New open 8B SOTA on ChartQA.

Fine-tune of Qwen3-VL-8B-Thinking. Slightly below Instruct variant on ChartQA.

Compact 8B VLM using LLM-based vision encoder outperforms Qwen3-VL 8B (89.6) and InternVL3.5 8B (86.7) on ChartQA.

Fine-tune of MiMoVL 7B. Strong ChartQA for 7B class.

Best chart understanding among evaluated models. End-to-end OCR with native chart reasoning.

Surpasses Gemini 2.5 Pro on ChartQA; 8B model trained with Gaussian GRPO.

Introduces Gaussian GRPO (G2RPO), replacing standard linear scaling in GRPO with non-linear distributional matching. OpenVLThinkerV2-7B achieves new SOTA for open-source 7B models on MMMU (71.6%), Mat

3B model with query-driven spatial priors outperforms many 8B models on text-heavy tasks.

8B open-weight multimodal model trained with GRPO+GDPO. Competitive with Gemini 2.5 Pro on DocVQA and chart understanding. New SOTA for open-weight VLMs on MMMU.

2B model matches OpenVLThinkerV2 8B (87.4) and beats Gemma E2B on ChartQA using LLM-based vision encoder.

4B end-to-end OCR model that ranks #1 on OmniDocBench among end-to-end models. Introduces "Layout-as-Thought" for structured layout representations. Outperforms Qwen3-VL-4B on ChartQA (+4.8) and Chart

8B open-weight multimodal model trained with GRPO+GDPO. Competitive with Gemini 2.5 Pro on DocVQA and chart understanding. New SOTA for open-weight VLMs on MMMU.

Layout-aware document reasoning model. Achieves 82.5 on ChartQA via structured Visual-Semantic Chain reasoning with layout priors.

Qwen3-VL-2B backbone with spatial pretraining. Strong for 2B class.

4B end-to-end OCR model that ranks #1 on OmniDocBench among end-to-end models. Introduces "Layout-as-Thought" for structured layout representations. Outperforms Qwen3-VL-4B on ChartQA (+4.8) and Chart