Top picks for Diagram Extraction (2026)
Reading flowcharts, org charts, architecture diagrams. Ranked from 452 live models on the OpenRouter catalog, weighted for vision input, structured output, reasoning quality.
Updated 2026-10-13 · prices checked at this morning's rebuild
Which should you use? Anthropic: Claude Opus 4.7 (batch) tops this ranking on blended score. If cost drives the decision, OpenAI: o3 (batch) is the cheapest of the leaders at $1.00/M input.
| # | Model | Score | In / 1M | Out / 1M | Context | |
|---|---|---|---|---|---|---|
| 1 | Anthropic: Claude Opus 4.7 (batch)anthropic/claude-opus-4.7:batch | 141 | $2.50 | $12.50 | 1,000,000 | Details → |
| 2 | OpenAI: o3openai/o3 | 139 | $2.00 | $8.00 | 200,000 | Details → |
| 3 | OpenAI: o3 (batch)openai/o3:batch | 139 | $1.00 | $4.00 | 200,000 | Details → |
| 4 | OpenAI: GPT-5openai/gpt-5 | 138 | $1.25 | $10.00 | 400,000 | Details → |
| 5 | OpenAI: GPT-5 (batch)openai/gpt-5:batch | 138 | $0.62 | $5.00 | 400,000 | Details → |
| 6 | Anthropic: Claude Sonnet 4.6anthropic/claude-sonnet-4.6 | 138 | $3.00 | $15.00 | 1,000,000 | Details → |
| 7 | Anthropic: Claude Sonnet 4.6 (batch)anthropic/claude-sonnet-4.6:batch | 138 | $1.50 | $7.50 | 1,000,000 | Details → |
| 8 | OpenAI: o4 Miniopenai/o4-mini | 137 | $1.10 | $4.40 | 200,000 | Details → |
| 9 | OpenAI: o4 Mini (batch)openai/o4-mini:batch | 137 | $0.55 | $2.20 | 200,000 | Details → |
| 10 | Anthropic: Claude Opus 4.7anthropic/claude-opus-4.7 | 136 | $5.00 | $25.00 | 1,000,000 | Details → |
| 11 | Google: Gemini 2.5 Progoogle/gemini-2.5-pro | 135 | $1.25 | $10.00 | 1,048,576 | Details → |
| 12 | Google: Gemini 2.5 Pro (batch)google/gemini-2.5-pro:batch | 135 | $0.62 | $5.00 | 1,048,576 | Details → |
| 13 | Google: Gemini 2.5 Flashgoogle/gemini-2.5-flash | 134 | $0.30 | $2.50 | 1,048,576 | Details → |
| 14 | Google: Gemini 2.5 Flash (batch)google/gemini-2.5-flash:batch | 134 | $0.15 | $1.25 | 1,048,576 | Details → |
| 15 | Anthropic: Claude Opus 5.5 (batch)anthropic/claude-opus-5.5:batch | 134 | $2.00 | $10.00 | 1,000,000 | Details → |
How we ranked these
For Diagram Extraction, we weight models on vision input, structured output, reasoning quality. Scores combine each model's public specs with independent benchmark results (Aider Polyglot coding scores, Artificial Analysis intelligence/coding/agentic indices) and live pricing. See full methodology →
About Diagram Extraction
Diagram extraction is the task of reading visual flowcharts, organizational hierarchies, system architecture diagrams, and similar structured graphics to extract their logical content, relationships, and text. You need this when you have hundreds of legacy diagrams in image format that must become machine-readable data, or when you're building systems that ingest visual workflows at scale. Good models recognize nodes, edges, labels, and hierarchy without hallucinating missing connections or misreading text overlaps common in dense diagrams. Poor models confuse similar shapes, lose spatial relationships, or fail on hand-drawn or low-contrast inputs. Processing time scales with image resolution and diagram complexity-a 4K screenshot of a 50-node architecture diagram can take 10-15 seconds on slower inference pipelines.
When to use: Use this when you have visual diagrams (flowcharts, org charts, network diagrams, system architecture) that you need to convert into structured data, searchable text, or editable formats, without manually redrawing them yourself.
Common questions
What is the difference between diagram extraction and general OCR?
OCR reads isolated text; diagram extraction must also understand spatial relationships, node connections, hierarchy, and the semantic meaning of shapes and arrows. A good diagram extractor knows that an arrow pointing downward in a flowchart means "flows to," not just "text appears below text." Models like Claude 3.5 Sonnet or GPT-4V handle this contextually, while basic OCR tools cannot.
How much does it cost to extract diagrams from a large document set, and how fast does it run?
API-based models (GPT-4V, Claude) typically cost $0.01-0.10 per diagram depending on resolution and provider pricing. Speed ranges from 2-30 seconds per image depending on model and queue load; batch processing APIs are slower but cheaper per unit. If you process 1,000 diagrams monthly, expect $10-100 in API costs plus time overhead.