Benchmark Radar™
Star —

CharXiv-D Benchmark

Explore score history ↗

CharXiv-D is the descriptive questions subset of the CharXiv benchmark, designed to assess multimodal large language models' ability to extract basic information from scientific charts. It contains descriptive questions covering information extraction, enumeration, pattern recognition, and counting across 2,323 diverse charts from arXiv papers, all curated and verified by human experts.

Modality
multimodal
Evaluates
multimodal, reasoning, structured output, vision
Openness
unknown
Importer
LLM Stats

Source provenance

CharXiv-D results and reported scores

Reported benchmark scores · LLM Stats

Reported scores: 17

ModelOrganizationReported valueReportedEvidence
Command A+Cohere0.882026-05-20evidence
GPT-4.1OpenAI0.8792025-04-14evidence
GPT-4.1 miniOpenAI0.8842025-04-14evidence
GPT-4.1 nanoOpenAI0.7392025-04-14evidence
GPT-4.5OpenAI0.92025-02-27evidence
GPT-4oOpenAI0.8532024-08-06evidence
North Micro Vision InstructCohere0.62026-08-12evidence
Qwen3 VL 30B A3B InstructQwen0.8552025-09-22evidence
Qwen3 VL 30B A3B ThinkingQwen0.8692025-09-22evidence
Qwen3 VL 32B InstructQwen0.9052025-09-22evidence
Qwen3 VL 32B ThinkingQwen0.9022025-09-22evidence
Qwen3 VL 4B InstructQwen0.7622025-09-22evidence
Qwen3 VL 4B ThinkingQwen0.8392025-09-22evidence
Qwen3 VL 8B InstructQwen0.832025-09-22evidence
Qwen3 VL 8B ThinkingQwen0.8592025-09-22evidence
Seed 2.1 ProByteDance0.9552026-06-24evidence
Seed 2.1 TurboByteDance0.9462026-06-24evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.

Which sources cite CharXiv-D?

Related multimodal, reasoning, structured output, vision benchmarks