RETIRED MAY 13 2026LEGACY QWEN 2.5131K CONTEXTTEXT + VISIONBATCH -50%
Qwen2.5 VL 72B Instruct API Pricing
Qwen 2.5 VL 72B Instruct was retired by Alibaba (Qwen) on May 13, 2026 - the qwen2.5-vl-72b-instruct model id no longer answers. Its last listed rates were $2.80/M input and $8.40/M output, kept here for invoice checks and migration math. Checked against Alibaba Cloud's Model Studio retirement notice 1840.
Input - per 1M tokens
$2.80/M
Source Alibaba Cloud flat
Output - per 1M tokens
$8.40/M
Context 131K flat
Cache N/A
$2.80/M
Cache no dollar row not listed
Effective - agentic blend
$3.25/M
92/8 split - no cache
§ 01 / TERMINAL
Run the numbers.
Live calculator pre-loaded with current Qwen2.5 VL 72B Instruct rates. Tweak spend, output mix, or cache assumptions and share the URL to share the calculation.
$ /mo
Workload split
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Real-world presets.
VISION
Invoice extraction
$0.024/doc
UI QA
Screenshot inspection
$0.042/screen
CHARTS
Dashboard explain
$0.020/image
MULTIMODAL RAG
Visual knowledge answer
$0.064/query
Calibrated · measured on the vendor's tokenizer · 2026-06-10 Auto-counts as you type
Counts use a chars-per-token calibration measured on the vendor's own published tokenizer (Qwen/Qwen3.5-397B-A17B, 2026-06-10). English prose is typically within a few percent; code and non-Latin scripts tokenize heavier. For billing-exact counts use the vendor's count-tokens API.
Characters 484
Words 75
Tokens (estimated) 93 tokens
Cost as input · uncached $0.00026 USD
Cost as output · uncached $0.00078 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Qwen 2.5 VL 72B Instruct Current | $2.80 | $8.40 | $3.25 agentic 92/8 | 131K | Legacy visual reasoning compatibility |
| Qwen3 VL Plus | $0.20 | $1.60 | $0.312 −10× | 256K | Current vision and document understanding |
| Qwen3 VL Flash | $0.05 | $0.40 | $0.078 deprecated · −42× | 256K | Cheap high-volume vision tasks |
| Qwen VL Max | $0.80 | $3.20 | $0.992 deprecated · −3.3× | 128K | Legacy Qwen vision flagship |
| Qwen VL Plus | $0.21 | $0.63 | $0.244 deprecated · −13× | 128K | Legacy low-cost vision apps |
| Qwen3 235B A22B | $0.70 | $2.80 | $0.868 deprecated · −3.7× | 131K | Current open MoE reasoning baseline |
| GPT-5.4 mini | $0.75 cache $0.075 | $4.50 | $0.541 −6.0× | 400K | OpenAI mini coding and CUA |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 −6.8× · promo | 1M | Google long-context Flash workloads |
Frequently asked.
Practical pricing questions, separated from calculator assumptions and regional tiers.
Q · 01 Can I still call Qwen 2.5 VL 72B Instruct? +
No. Alibaba (Qwen) retired
qwen2.5-vl-72b-instruct on May 13, 2026, and requests to it now fail. The rates on this page are the last ones Alibaba (Qwen) listed, kept for reading old invoices and for estimating a migration.Q · 02 What is Qwen2.5 VL 72B Instruct priced at? +
Qwen2.5 VL 72B Instruct is listed at
$2.8/M input and $8.4/M output in Alibaba Cloud Model Studio's International/Singapore deployment section. The page stores USD per-million-token pricing.Q · 03 What replaced Qwen2.5 VL 72B Instruct? +
Alibaba's retirement notice pointed Qwen2.5 users to the Qwen3.6 series. For vision work, compare Qwen3 VL Plus or Qwen3.6 Flash; Qwen3 VL Flash, the other suggestion here, goes offline on October 10, 2026.
Q · 04 Does this page use International or Global pricing? +
This page uses Alibaba Cloud Model Studio International deployment pricing, where endpoint and data storage are in Singapore and inference resources are dynamically scheduled globally excluding Chinese Mainland. Global, US, EU, China (Hong Kong), and Chinese Mainland sections can list different prices.
Q · 05 Is prompt caching priced separately? +
Alibaba marks context-cache support on some Qwen rows, but this row does not publish a concrete cache-read dollar price in the pricing table. The calculator therefore treats cached input as the same
$2.8/M baseline instead of inventing a discount.Q · 06 How is the effective price calculated? +
AI//COST uses the same 92/8 agentic blend everywhere. With no separate cache-read price published for this row, Qwen2.5 VL 72B Instruct's effective blended cost is
$3.25/M.Q · 07 Is there a Batch Invocation discount? +
Alibaba documents Batch Invocation as 50% off real-time input and output tokens for supported Qwen rows. The quote tiles show real-time list pricing; batch economics should be treated as a separate calculator variant.
Q · 08 Does Alibaba include a free quota? +
Many International Model Studio rows include a 1 million token free quota that is valid for 90 days after activating Model Studio. Free-quota eligibility is deployment- and model-specific, so production estimates should use the paid list prices shown here.