Cheapest vision model — image input without a flagship bill
Vision billing is easy to underestimate: images tokenize into large input chunks. The cheapest vision model is the cheapest row that actually accepts images and reads your screenshots well enough — not the cheapest text-only model.
On model pages, check modality badges for image input. Compare blended text prices as a lower bound, then read the provider’s image-token rules. Escalate to a flagship only for hard OCR or dense charts.
Popular frontier models by quality
Live OpenRouter pricing and Arena quality, cached about an hour. Not a static blog table.
| Model | Provider | Blend $ / 1M | Quality | Context |
|---|---|---|---|---|
| Claude Opus 5 | Anthropic | 15.00 | 93.4 | 1,000,000 |
| Claude Fable 5 | Anthropic | 30.00 | 91.1 | 1,000,000 |
| Qwen3.8 27B | Qwen | 1.71 | 89.0 | 1,000,000 |
| Gemini 3.7 Flash | 2.25 | 88.4 | 1,048,576 | |
| DeepSeek V4 Pro 0423 | DeepSeek | 1.14 | 88.2 | 1,024,000 |
Text price is not vision price
A model at $0.10/M text tokens can still bill a 2MB screenshot as thousands of tokens. Resize images, crop, or OCR first. The monthly cost table on this site is text-token based; multiply by a realistic image token estimate for vision jobs.
OCR vs VLM
If the job is “read this invoice,” a dedicated OCR pipeline plus a cheap text LLM can beat a vision flagship. If the job is “what’s wrong with this UI screenshot,” you need a VLM.
Gemini, GPT, Claude, open VLMs
All major labs ship vision SKUs; open VLMs appear as hosted rows too. Compare modality fields and price. Quality index is still text-arena Elo — it will not rank chart reading. Use the arena only when the replay includes images; otherwise test yourself.
Batch document understanding
For millions of pages, look at specialized document APIs. A general VLM is the flexible default, not always the cheapest factory line.
Practical shortlist
Find two image-capable rows, one cheap and one flagship. Run 20 real images. If the cheap model is ≥90% as good, route default there. Keep the flagship for low-confidence cases.
FAQ
What is the cheapest model that can see images?
- Search the leaderboard for vision/image modalities and sort by cost. Confirm image input on the model page badges.
Why is my vision bill so high?
- Image tokens. Downscale screenshots; don’t send 4K photos of a receipt.
Is Gemini cheaper for vision than GPT?
- Often flash-class Gemini is priced for volume, but check the current rows and image token policy.
Do open-weight VLMs work?
- Yes for many tasks. Compare hosted price vs self-host GPU cost.
Related guides
Also try: LLM leaderboard, compare tool, live arena.