21.9B- input
- $0.42
- output
- $1.25
- Context
- 123Ktokens
- Listed
- 1y ago
- Size
- 424Bparameters
Cost/1M tokens
Baidu's ERNIE-4.5-VL-424B-A47B is a post-trained vision-language model in the ERNIE 4.5 family, launched with an open-source release under Apache 2.0. Baidu says a heterogeneous Mixture-of-Experts design, with shared and modality-specific parameters, strengthens multimodal understanding without compromising text performance, and reports 47% Model FLOPs Utilization in pre-training its largest language model. It is built for visual-language understanding, with thinking and non-thinking modes for visual perception, document and chart understanding, and multimodal reasoning.
- Multimodal model
- Open weights · 424B
- Accepts images
- Reasoning
How it compares
Additional source evidence
- Source model
- baidu/ernie-4.5-vl-424b-a47b
- Source family
- ernie
- Last updated
- 2025-06-30
- Knowledge cutoff
- 2025-03-31
- Model context
- 123K
- Model output limit
- 16K
- Open weights
- Yes
- File attachments
- Yes
- Reasoning
- Yes
- Tool calling
- No
- Temperature control
- Yes
Complete source record (JSON)
{
"id": "baidu/ernie-4.5-vl-424b-a47b",
"name": "ERNIE 4.5 VL 424B A47B ",
"description": "Multimodal reasoning model for visual analysis, planning, and tool use",
"family": "ernie",
"last_updated": "2025-06-30",
"knowledge": "2025-03-31",
"attachment": true,
"reasoning": true,
"tool_call": false,
"temperature": true,
"modalities": {
"input": [
"image",
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 123000,
"output": 16000
},
"release_date": "2025-06-30",
"open_weights": true
}Frequently asked questions
Running locally
ERNIE 4.5 VL 424B A47B needs an estimated 295.8 GB of VRAM at Q4, 518 GB at Q8 and 974.6 GB at FP16. All figures are estimates from the 424B parameter count, covering weights plus about 15% and 0.5 GB for a short context.
No. Based on estimates, it does not fit on a 24 GB, 16 GB, 12 GB or 8 GB GPU at Q4, Q8 or FP16.
Yes, at Q4 it needs an estimated 295.8 GB. Assuming macOS lets the GPU use about 75% of unified memory, the smallest common Mac memory size that fits is 512 GB (estimate).
API and pricing
$0.42 per 1M input tokens and $1.25 per 1M output tokens. Paid providers tracked: 0.
A typical request of 10,000 input tokens and 1,000 output tokens costs $0.00545 at the listed price.
Privacy and data
Yes, ERNIE 4.5 VL 424B A47B has open weights, so it can be self-hosted for full control of your data.
Capabilities and specs
The context window is 123K tokens (123,000) and the maximum output is 16K tokens (16,000).
It reads images and text. Audio, video and PDFs are not supported input modalities.
No, tool calling (function calling) is not supported.
Yes, a reasoning mode is supported, but it is not always on.
The knowledge cutoff is 2025-03-31.
Quality and benchmarks
Qwen3 VL 235B A22B Instruct by Qwen with a blended price of $0.6325 per 1M tokens, Qwen3 Coder 480B A35B by Qwen at $0.475 per 1M tokens and GLM 4.5 Air by Z.ai at $0.425 per 1M tokens, matched on model kind, blended price and release date.
Qwen3 VL 235B A22B InstructERNIE 4.5 VL 424B A47B vs Qwen3 VL 235B A22B InstructQwen3 Coder 480B A35BERNIE 4.5 VL 424B A47B vs Qwen3 Coder 480B A35BGLM 4.5 AirERNIE 4.5 VL 424B A47B vs GLM 4.5 Air
Licensing and access
It is open weights, released under the Apache 2.0 license.
Yes, it is licensed under Apache 2.0, a permissive license that allows commercial use.
The weights are available from the Hugging Face repository baidu/ERNIE-4.5-VL-424B-A47B-PT.
Basics
ERNIE 4.5 VL 424B A47B was made by Baidu and first listed on OpenRouter on June 30, 2025.
No, the listed API price is $0.42 input and $1.25 output per 1M tokens, though the weights are open.
Generated from Vertical Tabs data, updated November 9, 2026.
Discussion
No comments yet
Be the first to share your thoughts.