Checked against the OpenRouter public models feed. Prices are copied, not estimated.

The Best Model ForWhich AI model to use, from ten labs.

The best AI model for image understanding

Ask a question about a photo. This is not image generation.

Short answer

Start with Gemini 4 Neon ($3.00 in · $15.00 out per million tokens). Neon accepts images, and also audio and video if the photo is part of a larger clip. Use it when the picture is one piece of a messier input. A cheaper stop is Gemini 4 Hydrogen ($1.50 in · $7.50 out).

Quality

Gemini 4 Neon

Google (Gemini) · $3.00 in · $15.00 out

Neon accepts images, and also audio and video if the photo is part of a larger clip. Use it when the picture is one piece of a messier input.

Gemini 4 Neon is listed at $3.00 input and $15.00 output per million tokens, with a 1,048,576-token context window.

Balanced

Nova 2 Lite

Amazon (Nova) · $0.30 in · $2.50 out

Nova 2 Lite is the reviewed everyday model for a receipt, a label, or a screenshot, at a lower rate than the flagships.

Nova 2 Lite is listed at $0.30 input and $2.50 output per million tokens, with a 1,000,000-token context window.

Cheapest sensible

Qwen3.8 Flash

Alibaba (Qwen) · $0.15 in · $0.47 out

Qwen3.8 Flash is the cheap open-weight option that lists image input. Test it on your actual photos.

Qwen3.8 Flash is listed at $0.15 input and $0.47 output per million tokens, with a 1,000,000-token context window.

Also on the feed

These match the factual limits of the task (modality, tools, context) and are not the reviewed picks. A row marked new has no guide note.

Questions

Which model should read a receipt?

Nova 2 Lite, Gemini 3.5 Flash Lite, or Qwen3.8 Flash. Ask for merchant, date, total, and tax. Throw away any line the photo does not show. A flagship is wasted unless the photo is genuinely hard.

Can these models create a new picture?

No. This task is reading. Picture-making is a different list of models, under Making a picture.