Claude Opus 5.5 is the best AI for image analysis. MiMo-V2.6-Pro comes close for 7% of the cost, and DeepSeek V4.1 Flash answers fastest.
Claude Opus 5.5 is the best AI for image analysis. MiMo-V2.6-Pro comes close for 7% of the cost, and DeepSeek V4.1 Flash answers fastest.
27 models · Updated September 2026 · Data from Artificial Analysis
Claude Opus 5.5
Anthropic · High effort
The highest score of any model that accepts images at this setting.
MiMo-V2.6-Pro
Xiaomi · Standard effort
86% of Claude Opus 5.5's score for 7% of the cost.
DeepSeek V4.1 Flash
DeepSeek · Max effort
The quickest full answer from a model scoring 38 or more.
Over 45 s at every tested setting
Claude Opus 5.5
Anthropic · High effort
The highest score of any model that accepts images at this setting.
MiMo-V2.6-Pro
Xiaomi · Standard effort
86% of Claude Opus 5.5's score for 7% of the cost.
DeepSeek V4.1 Flash
DeepSeek · Max effort
The quickest full answer from a model scoring 38 or more.
Each job favors a different model. Pick the one closest to yours.
Charts, diagrams and screenshots that need reasoning
Use
Claude Opus 5.5
Anthropic · High effort
Reads a chart and explains what changed. At volume, GPT-6 Sol scores 44.1 for $14.50 per 1,000 images against $39.41.
Charts, diagrams and screenshots that need reasoning
Use Claude Opus 5.5
Reads a chart and explains what changed. At volume, GPT-6 Sol scores 44.1 for $14.50 per 1,000 images against $39.41.
Describing and tagging photos in bulk
Use MiMo-V2.6-Pro
$2.73 per 1,000 images. It's MIT-licensed, so you can also host it yourself. Answers take about 61 seconds, which suits batch jobs better than live ones.
Live apps where someone waits for the answer
Use DeepSeek V4.1 Flash
A full answer in 12 seconds, against 41 seconds for Claude Opus 5.5. It scores 39.5.
Simple captions on a tight budget
Use GPT-6 Luna
$0.93 per 1,000 images, the cheapest paid option. It scores 33.9, so keep it to descriptions and tags.
Text in scans, receipts and handwriting
Text extraction is priced per page and often needs less reasoning, so it has its own page with costs and picks for scans and receipts.
PDFs, invoices and multi-page documents
Long documents add context-window and structured-output needs on top of vision.
Each dot is a model. Higher scores better, further left is cheaper, so the best deals sit top left. The cost axis is logarithmic: every step left is a big saving. Hover a dot for its numbers.
Sort by the pillar you care about, and set your monthly volume to see the bill. Costs include each model's thinking tokens at the setting shown.
Gemini 3.8 Flash. It scores 40.9 to Gemini 3.1 Pro's 29.7 and gives a full answer in 24 seconds instead of 35 seconds.
It isn't cheaper in practice. Its list price is lower ($0.75/$3.75 per million tokens against $2/$12), but it thinks far more, so 1,000 images cost about $19.75 against $17.07.
If speed matters more than the last point of quality, Gemini 3.7 Flash at medium effort answers in 7 seconds for $11.70 per 1,000 images and scores 39.6.
Every model on this page accepts image input. We compare them at the setting people actually run, on intelligence, real cost and time to answer.
Most models let you choose how long they think, and benchmark headlines use the maximum. That setting is slow and expensive: Claude Opus 5.5 gains 4 points at max effort but costs 3.3 times as much per task. GPT-6 Sol at max takes about 2.8 minutes per answer. Effort labels don't line up across labs either, so we don't match them by name. For each model we use its strongest setting that gives a full answer within 45 seconds in Artificial Analysis's tests. Switch to Max effort above for the benchmark numbers.
Intelligence is the Artificial Analysis Intelligence Index at that setting. Artificial Analysis doesn't publish a vision-only score, so this measures general reasoning. It predicts how well a model explains a chart or a screenshot, and says less about fine detail such as small print.
Cost starts from a typical job: 3,600 input tokens for the image and your prompt, and 2,400 output tokens. We then scale the output by how much each model actually wrote, thinking included, when Artificial Analysis tested it at that setting. A model that thinks twice as much as the median pays for twice the output. Where a model hasn't been measured, we use its list price and mark it.
Speed is the median time to a full answer in the same tests, thinking included.
Best value is the model with the most intelligence per dollar, and fastest is the one with the quickest full answer. Both only count models that score 38 or more, within 30% of the leader, and skip models their maker has replaced with a newer version.
Claude Opus 5.5 at high effort. It scores 53.6 on the Artificial Analysis Intelligence Index, the highest of any model that accepts images at everyday settings, and costs about $39.41 per 1,000 images. For bulk work, MiMo-V2.6-Pro scores 46.3 for $2.73. For live apps, DeepSeek V4.1 Flash answers in 12 seconds.
Gemini 3.8 Flash. It scores 40.9 to Gemini 3.1 Pro's 29.7 and answers in 24 seconds instead of 35 seconds. Because it thinks more, 1,000 images cost about $19.75, against $17.07 for Gemini 3.1 Pro.
GPT-6 Luna, at about $0.93 per 1,000 images (list price $0.10/$0.50 per million tokens). It scores 33.9, so it handles descriptions and tags better than charts that need reasoning. Qwen 2.5 VL 72B is free on OpenRouter, but it's an older model with rate limits on the free tier.
Between about $0.93 and $89.68, depending on the model and how long it thinks. Claude Opus 5.5 costs $39.41, GPT-6 Sol $14.50 and MiMo-V2.6-Pro $2.73. The estimate assumes a typical image and prompt, and scales each model's output by how much it actually wrote in testing.
Claude Opus 5.5, the highest-scoring model that accepts images at everyday settings (53.6 on the Artificial Analysis Intelligence Index). Every model on this page is a vision model, so the ranking above doubles as a vision model comparison.
GPT-6 Luna, at about $0.93 per 1,000 images including its thinking tokens. It scores 33.9. For a much stronger model that's still cheap, MiMo-V2.6-Pro scores 46.3 for $2.73.
Yes, for cost and wait time more than quality. Max effort adds a few points on benchmarks but often multiplies the cost and can push answers past two minutes. We rank every model at its strongest setting that answers within 45 seconds, and you can switch the page to max effort to compare.
Often, but not always. Pulling text out of an image depends on fine detail more than reasoning, and no public benchmark measures that for every model at everyday settings. Our OCR page covers cost per page and speed for text extraction, and suggests testing on a sample of your own documents.
Answer three questions and get a pick for each part of your workload, with the monthly cost.