Gemma 4 E4B vs Qwen3.5-4B on document tasks: Qwen wins the benchmarks, but the sub-scores tell a different story

Results live here: https://www.idp-leaderboard.org/ Ran both through the IDP Leaderboard (OlmOCR Bench, OmniDocBench, IDP Core) and the headline numbers aren't the interesting part. Top-line scores: Benchmark Gemma 4 E4B Qwen3.5-4B OlmOCR 47.0 75.4 OmniDoc 59.7 67.6 IDP Core 55.0 74.5 Qwen wins all