Skip to content
CostPerPrompt

Qwen2.5 VL 72B Instruct API Pricing

Alibaba (Qwen) · context window 128K · prices updated 2026-08-02

Input per 1M tokens
$0.8
Output per 1M tokens
$1.00
Cached input per 1M
$0.4
50% cheaper than fresh input

What real workloads cost on Qwen2.5 VL 72B Instruct

Workload Per request Per month
1K short requests/day (500 in / 150 out) $0.0006 $16.50
Support chatbot, 500 convs/day (~5K in / 1.4K out per conv) $0.0054 $81.00
Document pipeline, 10K docs/day (3K in / 500 out) $0.0029 $870

Model your exact traffic in the API cost calculator — it preloads Qwen2.5 VL 72B Instruct with caching and batch options.

Cheaper alternatives

Qwen3 Coder Flash Alibaba (Qwen) $0.195 in · $0.975 out DeepSeek V3.1 DeepSeek $0.25 in · $0.95 out Qwen3 14B Alibaba (Qwen) $0.2275 in · $0.91 out

More Alibaba (Qwen) models

Qwen3.7 Max Qwen3.6 Max Preview Qwen3 Max Thinking Qwen3 Max

Frequently asked questions

› How much does the Qwen2.5 VL 72B Instruct API cost?

Qwen2.5 VL 72B Instruct costs $0.8 per million input tokens and $1.00 per million output tokens, with cached input at $0.4 per million (50% cheaper). In practice, a typical short request (500 tokens in, 150 out) costs about $0.0006.

› What does 1,000 requests per day cost on Qwen2.5 VL 72B Instruct?

At 500 input + 150 output tokens per request, 1,000 requests/day costs about $16.50 per month. Heavier workloads scale linearly — use our API cost calculator to model your exact traffic.

› How can I reduce Qwen2.5 VL 72B Instruct costs?

Three levers, in order of impact: (1) route routine requests to a cheaper tier and keep Qwen2.5 VL 72B Instruct for hard cases; (2) enable prompt caching — cached input is 50% cheaper, which typically halves chatbot bills; (3) use the batch API for non-urgent jobs at a ~50% discount.

› Is Qwen2.5 VL 72B Instruct worth the price?

It depends on task difficulty. Flagship models earn their premium on complex reasoning, coding, and high-stakes output; mini/flash tiers handle classification, extraction and routine chat at 10–30× lower cost. Benchmark a sample of your real workload on both before committing.