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
- Grok Bot Ran a 12-Hour Data Analysis Nonstop, Handling API Rate Limits Itself — Daniel_Farinax · 2026-10-16
- One year of Recursive Language Models: GPT-5-mini beats GPT-5 by 110% on 132k-token tasks, cheaper too — a1zhang · 2026-10-16
- Users Report Claude Models Suddenly More Compliant After Forced Compaction — RealSharpNinja · 2026-10-16
- Prediction: Opus 5.5-level models on laptops by 2027, backed by Qwen 27B data — cephaloform · 2026-10-16
- Users Say Krea 2 Output Degraded Sharply Since Krea 3 Beta Launched — szywis · 2026-10-16
- LLM-as-a-Verifier lifts Nemotron 3.5 Super by 9-10 points on agentic benchmarks — simonguozirui · 2026-10-16