Model reference · Synced 2025-04-29
Qwen2.5-VL 72B Instruct
Qwen2.5-VL 72B Instruct is an AI model from OpenRouter. 33K context window. Capabilities: tool calling, multimodal vision, open weights. Available on 7 providers. Cheapest listing: $0 input / $0 output per 1M tokens.
Quick facts
- Cheapest input: $0 per 1M tokens (OpenRouter)
- Cheapest output: $0 per 1M tokens
- Context window: 33K tokens
- Max output: 8K tokens
- Release date: 2025-02-01
- Knowledge cutoff: 2024-10
- Capabilities: tool calling, multimodal vision, open weights
- Provider count: 7
Provider pricing
Same model, different providers, different prices. Cheapest first.
| Provider | Input / 1M | Output / 1M | Context | Listed |
|---|---|---|---|---|
| OpenRouter | $0 | $0 | 33K | 2025-02-01 |
| Nebius Token Factory | $0.25 | $0.75 | 128K | 2025-01-20 |
| NovitaAI | $0.8 | $0.8 | 33K | 2025-03-25 |
| OVHcloud AI Endpoints | $1.01 | $1.01 | 33K | 2025-03-31 |
| Alibaba (China) | $2.29 | $6.88 | 131K | 2024-09 |
| Alibaba | $2.8 | $8.4 | 131K | 2024-09 |
| LLM Gateway | $2.8 | $8.4 | 131K | 2024-09 |
Prices synced daily from models.dev + provider docs.
How to use this model
If you're picking Qwen2.5-VL 72B Instruct for a project, the three things that matter most:
- Compare it side-by-side with one or two alternatives in the live comparison tool. Pricing differences matter more than benchmarks at scale.
- Pick the cheapest provider that meets your latency / SLA need. Big spread across providers for the same weights.
- Re-evaluate every 3 months. Frontier prices drop fast; a model that's cheapest today may not be in a quarter.
Related models
FAQ
How much does Qwen2.5-VL 72B Instruct cost? $0 input / $0 output per 1M tokens at the cheapest listing. See the table above for other providers.
What is the context window? 33K tokens.
Which providers offer it? Alibaba, OpenRouter, Alibaba (China), NovitaAI, Nebius Token Factory, OVHcloud AI Endpoints, LLM Gateway.
Where do these numbers come from? models.dev + provider documentation, synced daily. About the data.