October 2026 Model Upgrade: Faster, More Accurate, and API Updates
The new model is now the default across our entire platform. For API integrations, requests without a version parameter automatically run on 261029, while developers needing previous output behavior can explicitly pin to version 260428.
What’s New in Model Version 261029
Our latest model iteration focuses on resolving real-world document edge cases while shaving valuable seconds off each recognition task:
Inference latency has dropped substantially, reducing round-trip times by up to 35% on standard documents and accelerating bulk batches in the workspace.
Multi-tier header rows, borderless financial statements, and nested cells now parse with greater structural consistency into clean HTML, CSV, or XLSX.
Sub-pixel clarity and damaged invoice printouts are handled with higher confidence, significantly reducing dropped digits and misread punctuation.
API Versioning: Defaulting to 261029, Pin 260428 When Needed
To provide immediate performance gains while preserving stability for existing workflows, we have updated how versioning operates across our Platform API:
Applied automatically to all calls to /ocr, /table, and /extract when no version parameter is passed. You do not need to alter your existing code to start benefiting from the new model.
If your downstream parsing logic, automated testing, or production pipelines depend strictly on the legacy model behavior, you can explicitly pin to it by passing ?version=260428.
Here is how you can use the default model or explicitly pin to the previous version:
# 1. Use the new 261029 model by default (no version parameter needed)
curl -X POST "https://freeocr.ai/api/v1/platform/ocr" \
-H "Authorization: Bearer ocr_your_api_key" \
-F "image=@document.png"
# 2. Or explicitly pin to the previous 260428 model
curl -X POST "https://freeocr.ai/api/v1/platform/ocr?version=260428" \
-H "Authorization: Bearer ocr_your_api_key" \
-F "image=@document.png"Deprecation Notice: Extraction Version 260215 Sunsets on November 11
With the rollout of 261029 as our standard engine, our oldest structured extraction version, 260215 (used on /api/v1/platform/extract), will be officially retired on November 11, 2026.
Zero Downtime: Seamless Internal Redirection
After November 11, any API request sent with version=260215 will continue to work without failing or breaking your code. Behind the scenes, requests will automatically alias and redirect to 260428.
Recommended Action: If your integration explicitly passes version=260215, please update your configuration to 260428 or upgrade directly to the default 261029 to take advantage of faster execution and sharper field extraction.
Next Steps
You can explore the full parameters and error responses in our API Documentation. For hands-on testing, try out our free web tools for OCR, Table Extraction, and Data Extraction.