EditReward Bench: Benchmark for Instruction-Guided Image Editing

EditScore provides a series of open-source reward models ranging from 7B to 72B parameters for evaluating instruction-guided image editing. The benchmark likely contains data used to train and evaluate these models, with the largest model reportedly surpassing GPT-5 on their internal benchmark. The dataset was last updated on October 17, 2025.

Use Cases

• Benchmarking image editing models based on reward model scores
• Training reward models for instruction-guided image editing
• Evaluating the alignment of edited images with textual instructions
• Comparing the performance of open-source and proprietary vision-language models on editing tasks

Strengths

• Includes state-of-the-art models ranging from 7B to 72B parameters
• The largest model reportedly surpasses GPT-5 on their internal benchmark
• Models are open-source and designed to match proprietary VLM performance

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count and file formats are unknown, which may limit suitability assessment

Provenance

Source

EditScore

Freshness

Last updated 2025-10-17 04:50:28