Cheapest Vision LLM in India (₹ Pricing)

If you need a model that can read images as well as text, Qwen3.8-27B is the cheapest kind of vision-capable option on unoblox's catalog: it is free (₹0 per token for input and output), as is Gemma 4 26B A4B.

Free vision-capable models on unoblox

Model	Input (₹ / 1M tokens)	Output (₹ / 1M tokens)	Notes
Qwen3.8-27B	₹0	₹0	Open weights, accepts image input
Gemma 4 26B A4B	₹0	₹0	Open weights, accepts image input

Because it is free, Qwen3.8-27B is the practical starting point if your workload needs to look at images — receipts, screenshots, product photos, scanned documents — and you're price-sensitive.

What "vision" means in practice

A vision-capable model accepts image content in the same request as your text prompt — you're not running a separate OCR step and feeding text back in; the model reads the image directly alongside your instructions. This is useful for tasks like describing an image, extracting structured data from a photographed document, or answering questions about a chart or screenshot.

If you need more capability than Qwen3.8-27B offers

Several of the larger closed models on the catalog are also generally multimodal and accept image input, at a higher per-token rate:

Model	Input (₹ / 1M tokens)	Output (₹ / 1M tokens)
GPT-4.1	₹202.46	₹809.82
Claude Sonnet	₹202.46	₹1,012.28
GPT-4o	₹253.07	₹1,012.28
Claude Opus	₹506.14	₹2,530.71

These are worth the extra cost when a task genuinely needs stronger reasoning over what's in the image, not just recognition — for example, reasoning about relationships between multiple charts rather than reading one clearly-labelled receipt. Always confirm image-input support and the exact rate on the model's own page before building against it.

A worked example

Image-heavy requests tend to be input-heavy, because an image encodes to a meaningful number of tokens before your text prompt is even counted. A request that works out to 4,000 input tokens (image plus prompt) and 300 output tokens costs nothing in token charges on Qwen3.8-27B, because the model is free. The same request on GPT-4o costs (4,000/1,000,000 × ₹253.07) + (300/1,000,000 × ₹1,012.28) ≈ ₹1.32, or roughly ₹13,160 for 10,000 such requests a month.

Calling a vision model from your code

The request shape is the same OpenAI-compatible format you already use for text-only calls — you add image content to the message, and the model field points at a vision-capable id.

base_url: https://api.unoblox.ai/v1
api_key: ub-gw-xxxxxxxxxxxxxxxxxxxx
model: qwen/qwen3.8-27b

Frequently asked questions

Is Qwen3.8-27B the only vision-capable model on unoblox? It's free (as is Gemma 4 26B A4B) and accepts image input. Larger models like GPT-4o, GPT-4.1, and the Claude Sonnet/Opus family are also generally multimodal — check each model's page to confirm before you build against it.

Does "open weights" mean anything for how I call Qwen3.8-27B? No — you call it through the same OpenAI-compatible endpoint and key as every other model on unoblox. "Open weights" describes how the model itself is distributed, not how you access it here.

Is image processing billed differently from text? No separate line item — an image is encoded into tokens the same way text is, and counted as part of your input tokens at the model's published input rate.

Can I send multiple images in one request? That depends on the model's own limits rather than anything unoblox adds on top. Check the model's page and your SDK's documentation for the exact constraints.

Does using a vision model mean my images are stored or used for training? That depends on the underlying model provider's own policy — check the relevant provider terms linked from the model's page rather than assuming either way.

Is there a free way to test a vision workload before paying? Yes — Qwen3.8-27B is free, so you can prototype image workloads at no token cost.

Get started in rupees → https://unoblox.ai/sign-in