Code OCR Test: DeepSeek Vision Compresses Tokens but Introduces Wording Errors
teortaxesTex · x · 2026-08-22
The author tested model OCR capabilities on code screenshots. DeepSeek-Vision compressed a 1000-word code screenshot into 330 tokens but introduced 3 nontrivial wording changes. The test suggests that models cannot be fully trusted for one-shot OCR replacement of high-density code screenshots unless converted to unambiguous rendering formats.
Related event: DeepSeek-Vision Compresses 1000-Word Images to 330 Tokens with Some Errors(2 posts)→
More from Models
- New Meta: Labs Trading Free Usage for Training Data — NathanpmYoung · 2026-08-22
- Speculation suggests strong MoE architecture balances inference speed with knowledge recall — jd_pressman · 2026-08-22
- User Tests New Model: Strong at Fermi Estimates and Strict Poetry Rewriting — jd_pressman · 2026-08-22
- VulcanBench: Grok 4.5 High Leads, Max Effort Doesn't Equal Better Accuracy — elonmusk · 2026-08-22
- Rumor: GLM 5.3 Flash Derived from Distillation and RL — teortaxesTex · 2026-08-22
- Grok Voice Model Tops New Speech Agent Arena Benchmark — XFreeze · 2026-08-22