Vision LLM API pricing in India

unoblox is OpenAI-API compatible, which means a vision-capable model on it accepts image inputs the same way OpenAI's own multimodal models do — as part of the same messages array, billed as part of the same per-token ₹ rate as text. There's no separate "vision surcharge" line; the image simply becomes some number of input tokens before the model ever sees it, and the exact count depends on the image's resolution and the specific model's vision encoder, which is why this page works in a worked example rather than a fixed per-image number.

What's confirmed vision-capable

qwen/qwen3.8-27b is unoblox's open-weight model explicitly listed with vision support, at ₹16.32 / ₹48.96 per 1M tokens (input/output). Several of the proprietary frontier families on unoblox — the GPT-4o/4.1/5 line and Claude's Opus/Sonnet line — are multimodal in their standard form and commonly used for image understanding as well; confirm exact modality support on each model's /models page before building against it, rather than assuming every listed model accepts images.

A worked ₹ example

Assume a workload sends about 1,000,000 mixed text-and-image input+output tokens a day at roughly a 2:1 input:output split — 0.67M input and 0.33M output tokens. That input figure already includes whatever tokens your images convert to; check a given model's documentation for how it tokenises a specific resolution before finalising your own estimate.

Model id	₹ / 1M input	₹ / 1M output	₹ / day at this volume
qwen/qwen3.8-27b	₹16.32	₹48.96	₹27.20
openai/gpt-4o	₹252.00	₹1,008.00	₹504.00
openai/gpt-4.1	₹201.60	₹806.40	₹403.20
openai/gpt-5	₹126.00	₹1,008.00	₹420.00

Claude Opus (anthropic/claude-opus-4-5/-4-8/-5, ₹504.00 / ₹2,520.00 per 1M tokens) would run ₹1,176.00/day at the same volume — useful context for how much a premium multimodal model costs relative to the open-weight option above it. For any vision-relevant model not shown here, check /models for its live rate rather than guessing.

Why resolution matters more than the model choice here

Because image tokenisation scales with resolution, downsizing images before sending them (when your use case tolerates it) reduces input tokens the same way trimming a text prompt would — and that reduction applies at whatever ₹ rate your chosen model bills, so it compounds with model choice rather than replacing it.

A minimal request

curl https://api.unoblox.ai/v1/chat/completions \
  -H "Authorization: Bearer ub-gw-xxxxxxxxxxxxxxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen/qwen3.8-27b","messages":[{"role":"user","content":"Describe this image."}]}'

Frequently asked questions

Is there a separate price for image inputs? No — images are billed as input tokens at the same per-1M-token ₹ rate as text, once converted to tokens by the model's vision encoder.

Which unoblox model is confirmed to support vision? qwen/qwen3.8-27b is explicitly listed as an open-weight vision model; check /models for current modality support on other listed models before relying on it.

Does a higher-resolution image always cost more? Generally yes, since more visual detail typically converts to more input tokens — the exact scaling depends on the model's vision encoder.

Is there a free way to test a vision integration? The fixed ₹0 model in the price list, qwen/qwen3-1.7b, is text-only — for vision testing, budget a small run on qwen/qwen3.8-27b and check actual token usage in your dashboard.

Can I mix text-only and vision calls under one key? Yes — the model is chosen per request; one unoblox key can call both text-only and vision-capable models.

How is all of this billed? In rupees, on a monthly GST invoice from an Indian entity, input tax credit claimable, same as every other model on unoblox.

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