Reduce yourOpenAIvision model spending by 60%+
What we do
We help you cut the cost of vision related requests to LLMs in your startup, by training or fine-tuning custom computer vision models for your use case.
Signup
Visual Question Answering
Gemini 1.5 Pro
Prompt: What is happening to the three cells in the center diagonal of this image? (short answer)
Response: The cells in the center diagonal are undergoing mitosis. This is evident from the visible chromosomes, spindle fibers, and breaking down nuclear envelope.
$0.00200 per request
or $1 for 500 requests
9.5 seconds
average response time
Custom Model
This model was trained using model distillation on a custom biology dataset.
Prompt: What is happening to the three cells in the center diagonal of this image? (short answer)
Response: The three cells are in different stages of mitosis: anaphase (top, chromosomes separating), prophase (middle, chromosomes condensing), and metaphase (bottom, chromosomes aligned in the center).
$0.00050 per request
or $1 for 2,000 requests
3.4 seconds
average response time
Image Captioning
GPT-4o (gpt-4o-2024-08-06)
Prompt: How many dogs are in this image? Output a JSON in the following format: {"dogs": 3}
Response: {"dogs": 7}
$0.00291 per request
or $1 for 344 requests
711 milliseconds
average response time
Custom Model
This model was trained using model distillation (from GPT-4o) to count the number of dogs in an image.
Response: {"dogs": 7}
$0.0001 per request
or $1 for 10,000 requests
104 milliseconds
average response time
Object Detection and Counting
GPT-4o (gpt-4o-2024-08-06)
Prompt: How many dogs are in this image? Output a JSON in the following format: {"dogs": 3}
Response: {"dogs": 7}
$0.00291 per request
or $1 for 344 requests
711 milliseconds
average response time
Custom Model
This model was trained using model distillation to count the number of dogs in an image.
Response: {"dogs": 7}
$0.0001 per request
or $1 for 10,000 requests
104 milliseconds
average response time
Pricing
Model Training
$32
- Compile a custom dataset to use during training
- Distill an existing model to create a application-specific model OR train a new model from scratch
We take on the risk: If we can't develop a cheaper model with the same accuracy within the next 14 days this fee will be refunded.
Inference API
Replicate cost + 10%
Our priority is on training accurate, cost-effective models, not managing inference. We use Replicate for deployment, offering a reliable API with top performance. A 10% markup on Replicate’s fees covers our operational costs.