Benchmarking 8 open OCR models: cost drops from $0.51 to $0.095 per 1,000 pages
vanstriendaniel · x · 2026-11-13
The author ran 8 open OCR models from the last 18 months on the same 850 pages, on a single GPU via Hugging Face Jobs. Cost fell from $0.51 to $0.095 per 1,000 pages — a near-straight line halving roughly every 10 months, though most of the drop comes from one new model: LightOnOCR-3-0.8B from LightOn.
Practical notes:
- Run the cheapest model on your own pages
- Everything was served with vLLM plus a batch tool to saturate throughput
- The author shares tooling to compare models on your own document collection
Related event: Open-Source OCR Cost Drops 5x in 18 Months, Benchmark Shows(2 posts)→
More from Models
- LMArena: top leaderboard scores move 3x faster while April's best model costs 20x less — arena · 2026-11-14
- Chinese Paper Maps CoT's Hidden Topology: Reasoning Behaves Like a Molecule, Not a Line — mikeflache · 2026-11-14
- 1.26M blind votes show users judge AI more harshly each month — models aren't declining — rohanpaul_ai · 2026-11-14
- DeepSeek 4.1 Flash Max Is First Open-Weight Model to Top 1% on Bug-Finding Benchmark — mariofilhoml · 2026-11-14
- Meta's unannounced Muse-Spark-1.4 model spotted in GitHub PR with real API usage logs — testingcatalog · 2026-11-14
- MathArena First Non-OpenAI #1: Gemini 4 Carbon Overtakes GPT-6.1 Sol — vedantmisra · 2026-11-14