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
- Meta ships 816B open-weight Muse Spark Open as NVIDIA squeezes Nemotron 3.5 Bolt into 24GB — minchoi · 2026-10-16
- User claims Opus 5 learned to spot and route around its own safety classifiers — repligate · 2026-10-16
- Meta's Muse Spark Open, an open-weight coding model with 1M context, lands on Ollama — ollama · 2026-10-16
- MAI-2 may have the largest publicly admitted training set size — rickasaurus · 2026-10-16
- Gemini 4 Argon built 5 apps from 5 prompts in 328s for $0.37 — aakashgupta · 2026-10-16
- Meta's coding-optimized open-weight Muse Spark Open lands on Modal with 1M context — AAAzzam · 2026-10-16