Guides

Multimodal & Vision

Cortiqa models can process both text and images in a unified request, enabling visual reasoning, document OCR, chart analysis, and user interface inspection.

Overview

You can provide images as base64-encoded strings or public URLs inside the message content array.

Dedicated Vision Architecture

The upcoming dedicated Falin Vision model (falin-vision, 128K context) will provide state-of-the-art native document perception, fine-grained diagram OCR, and spatial UI navigation.

Sending Images

vision_demo.py

1import base64
2from cortiqa import Cortiqa
3
4client = Cortiqa()
5
6with open("architecture_diagram.png", "rb") as f:
7    base64_image = base64.b64encode(f.read()).decode("utf-8")
8
9response = client.chat.completions.create(
10    model="openai/gpt-oss-120b",
11    messages=[
12        {
13            "role": "user",
14            "content": [
15                {"type": "text", "text": "Analyze this architecture diagram and highlight any single points of failure."},
16                {
17                    "type": "image_url",
18                    "image_url": {
19                        "url": f"data:image/png;base64,{base64_image}"
20                    }
21                }
22            ]
23        }
24    ]
25)
26
27print(response.choices[0].message.content)

Supported Formats

Format	MIME Type	Max Upload Size
PNG	`image/png`	20 MB
JPEG	`image/jpeg`	20 MB
WebP	`image/webp`	20 MB
GIF	`image/gif`	20 MB

Best Practices

Resolution: High resolution images (e.g. 1080p - 4K) yield much better text extraction for diagrams and spreadsheets.
Clarity: Crop irrelevant borders or margins before encoding to minimize token usage and latency.
Prompts: Formulate specific visual questions (e.g., “What is the value in row 3 column 2?” rather than “What is this?”).

Was this page helpful?

System Prompts

Tool Use