Top picks for Chart & Graph Reading (2026)
Pulling numbers off charts in research papers and reports. Ranked from 334 live models on the OpenRouter catalog, weighted for vision input, reasoning quality, structured output.
| # | Model | Score | In / 1M | Out / 1M | Context | |
|---|---|---|---|---|---|---|
| 1 | Anthropic: Claude Sonnet 4.6anthropic/claude-sonnet-4.6 | 160 | $3.00 | $15.00 | 1,000,000 | Details → |
| 2 | Anthropic: Claude Opus 4.7anthropic/claude-opus-4.7 | 158 | $5.00 | $25.00 | 1,000,000 | Details → |
| 3 | OpenAI: GPT-5.4openai/gpt-5.4 | 155 | $2.50 | $15.00 | 1,050,000 | Details → |
| 4 | Anthropic: Claude Opus 4.8anthropic/claude-opus-4.8 | 151 | $5.00 | $25.00 | 1,000,000 | Details → |
| 5 | Google: Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-preview | 151 | $2.00 | $12.00 | 1,048,576 | Details → |
| 6 | Google: Gemini 3.5 Flashgoogle/gemini-3.5-flash | 151 | $1.50 | $9.00 | 1,048,576 | Details → |
| 7 | MoonshotAI: Kimi K2.6moonshotai/kimi-k2.6 | 151 | $0.66 | $3.50 | 262,144 | Details → |
| 8 | MiniMax: MiniMax M3minimax/minimax-m3 | 149 | $0.30 | $1.20 | 1,048,576 | Details → |
| 9 | OpenAI: GPT-5.5openai/gpt-5.5 | 148 | $5.00 | $30.00 | 1,050,000 | Details → |
| 10 | MoonshotAI: Kimi K2.7 Codemoonshotai/kimi-k2.7-code | 148 | $0.61 | $3.07 | 262,144 | Details → |
| 11 | OpenAI: GPT-5.4 Miniopenai/gpt-5.4-mini | 147 | $0.75 | $4.50 | 400,000 | Details → |
| 12 | Qwen: Qwen3.6 Plusqwen/qwen3.6-plus | 146 | $0.33 | $1.95 | 1,000,000 | Details → |
| 13 | OpenAI: GPT-5.4 Nanoopenai/gpt-5.4-nano | 146 | $0.20 | $1.25 | 400,000 | Details → |
| 14 | Qwen: Qwen3.5 397B A17Bqwen/qwen3.5-397b-a17b | 146 | $0.39 | $2.45 | 256,000 | Details → |
| 15 | Qwen: Qwen3.7 Plusqwen/qwen3.7-plus | 146 | $0.32 | $1.28 | 1,000,000 | Details → |
How we ranked these
For Chart & Graph Reading, we weight models on vision input, reasoning quality, structured output. 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 Chart & Graph Reading
Chart and graph reading is the task of extracting numerical values, labels, and trends from visual data representations in research papers, reports, and presentations. You need this when manual data entry becomes a bottleneck or when you're processing dozens of documents where OCR alone fails on axis labels, legend items, and plotted values. Good models handle rotated text, overlapping data points, and logarithmic scales without hallucinating values; poor ones confuse similar colors, misread small fonts, and invent numbers that don't appear anywhere on the image. The main trade-off: vision models that extract text accurately often process images slowly (3-10 seconds per chart), while faster models skip annotation details that matter for downstream analysis.
When to use: Use this when you need to pull specific numbers from charts, graphs, or plots faster than a human could type them, especially when processing research papers, financial reports, or scientific datasets where manual transcription would take hours.
Common questions
What is the best AI model for reading charts and extracting numerical data?
GPT-4 Vision and Claude 3.5 Sonnet are currently the strongest general-purpose choices, with Claude 3.5 Sonnet showing slightly better accuracy on dense scientific plots. For specialized financial charts, specialized models trained on annual reports often outperform general vision models, but they cost more per image and require vendor lock-in.
How much does it cost to extract data from 1,000 charts, and does speed matter?
At current API pricing, GPT-4 Vision costs roughly $3-8 per 1,000 images depending on resolution; Claude 3.5 Sonnet runs $0.30-1.50 per 1,000. Speed matters if you're on deadline: batch processing takes 1-3 hours for 1,000 charts, while real-time extraction in a document pipeline needs sub-second response times, which typically requires local models or cached inference.