ThaiOCRBench: A Vision-Language Benchmark for Thai Text-Rich Images

2,808 human-annotated samples across 13 diverse tasks, including table parsing, chart understanding, and visual question answering. ThaiOCRBench is a benchmark for evaluating vision-language models on Thai text-rich visual understanding tasks, created by typhoon-ai and last updated in December 2025.

Use Cases

• Benchmarking model performance on Thai table parsing tasks based on the described task diversity.
• Evaluating chart understanding capabilities for Thai-language charts based on the benchmark's scope.
• Testing key information extraction from Thai documents based on the described task categories.
• Assessing full-page optical character recognition for Thai text based on the benchmark's design.

Strengths

• 2,808 human-annotated samples, providing a substantial evaluation set.
• Covers 13 diverse tasks, suggesting broad coverage of vision-language challenges.
• Specifically designed for the Thai language, addressing a less common language in benchmarks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known but the specific data format and file structure are unknown.
• Data may reflect geographic or source bias inherent to its collection method.

Provenance

Source

typhoon-ai on Hugging Face.

Collection Method

Human-annotated samples, likely gathered for benchmark creation.

Time Range

null

Freshness

Last updated 2025-12-02 08:28:45.

Geography

Focus on Thai language, suggesting primary relevance to Thailand.