Product Updates

​

APIs

• AI Scraper
  • Improved AI Scraper reliability, performance, and schema validation.
• Web Search
  • Improved web search results for Twitter-related queries.
• Deep-Research
  • Ensured Deep Research API responses consistently include their underlying sources.
• Embedding-v2
  • Added support for configurable embedding dimensions, with 4096 as the default for v2/embedding.
• Classification
  • Fixed unusual timeout issues with image-classification.
• VOCR
  • Enhanced VOCR word-level bounding boxes when using the fine_grained flag for more complete coverage.

​

DeepResearch (open-source)

• Migrated DeepResearch to OpenRouter for stable performance.
• Reduced end-to-end latency, making long runs noticeably faster.

​

APIs

• Prediction
  • Resolved intermittent timeout issues.
• Speech-to-Text
  • Fixed a bug that caused Javanese audio requests to fail.
  • Fixed an issue where m4a files failed to transcribe.
• VOCR
  • Improved error propagation for clearer, more actionable error details.
• Translate
  • Fixed a bug where valid requests were incorrectly blocked.
• Log ID
  • Each request now comes with a Log ID to make support and debugging easier.

​

INTERFAZE

• Fixed a bug where URLs returned were placeholders.
• Improved error propagation for clearer debugging.

​

APIs

• Prediction
  • Fixed timeout issues to improve reliability and response consistency.
• Speech-to-Text
  • Improved speaker label accuracy and diarization for better multi-speaker transcription.
  • Fixed an issue preventing STT from accessing certain URLs.
• Web Search
  • Refined handling of various edge cases to ensure more accurate results.
  • Expanded Historical Finance Index, now available for querying.

​

INTERFAZE

• Improved language detection accuracy for enhanced contextual understanding.
• Added support for Feature Context and Thinking chunks during streaming for deeper insights.
• Updated Playground UI to visualize these chunks, along with new documentation for developers.

​

APIs

• Web Search:
  • Enhanced AI Overview for richer, more accurate summaries
  • Improved historical finance data retrieval
• VOCR:
  • Fixed issues with random spikes in latency.
• Speech to Text (STT):
  • Fixed timeout issues for smoother response handling
• Summary:
  • Fixed timeout issues for smoother response handling
• Prompt Engine:
  • Resolved timeout issues and improved reliability
• Speech to Text:
  • We’ve improved our multilingual diarization service to reduce speaker confusion
• Translate:
  • Fixed and improved input language detection accuracy
• Classification
  • Upgraded multilingual models for higher-quality results across languages

​

SDKs

• Dropped Python support for versions 3.8 and below

​

INTERFAZE

• Enhanced the model’s vision capabilities
• Added support to generate structured JSON with AI directly in the Playground
• Improved error handling pipeline for better response consistency

​

APIs

• VOCR:
  • Fixed the issue where the lines and section output of ocr format was different when there is a prompt and no prompt. Fixed to standardize the format
  • Added confidence score for the words detected in the ocr format
• Web Search: Improved the speed and quality of the web search model significantly
• Speech to Text (STT):
  • Updated suppport for longer than 5+ hour audio clips with diariazation
  • Fixed input validation for empty audio files
• Image Translate:
  • Fixed input validation for the image translate API
  • Fixed a temporary outage caused by a recent deployment that affected image generation requests
• AI Scraper: Fixed issue of large broad element selection for scraping. Now supports things like “body” or “main” as the element prompts.

​

SDKs

• VOCR Added confidence score for the words detected in the ocr format

​

DOCS

• Updated JigsawStack Google Sheets integration
• Added confidence score for words detected in the VOCR API

​

Others

• Updated our Postman Collection to include the most up to date APIs and features

​

Fixes

• Fixed Embedding v1 API where sometimes the model would fail inconsistently
• Fixed Classification API where the model would fail inconsistently for certain images
• Fixed AI Scraper where the model would fail for pdf with large amount of pages

​

APIs

• Speech to Text (STT): Added confidnece score for the language detected. Available if language parameter is not provided or set to “auto”.
• Object to Detection:: Improved the speed of the object detection model
• Improved the speed of APIs that involves processing binary data. This includes:
  • Object Detection
  • Speech to Text (STT)
  • Embedding v1
  • Embedding v2
  • Image Translation
  • vOCR
  • NSFW
  • File Upload

​

DOCS

• Introduced JigsawStack Google Sheets integration
• Added confidence score for the language detected in the STT API
• Added warning for the deprecated Image Generation API

​

Fixes

• Updated Object Detection API to accept “object” parameter as an alternative to “object_detection”
• Fixed Object Detection API for pure GUI images
• Improved Speech to Text (STT) webhook error handling with 400 status code for empty files
• Enhanced Object Detection tag generation for more accurate and relevant results

​

SDKs

• Updated Object Detection now supports “object” parameter as an alternative to “object_detection”
• Updated Speech to Text (STT) to return language_detected and confidence in the response

​

DOCS

• Added documentation for Embedding v2 parameters: instructions and query
• Added documentation for Object Detection parameters: return_masks
• Fixed CURL example for File Uploads
• Fixed C# examples on all APIs
• Updated Speech to Text (STT) documentation to include parameters: chunk_duration, language_detected, and confidence

​

Fixes

• Added support for NSFW API when images are not available
• Added support for HTML to Any now works with all HTML tags as input
• Fixed File Upload now works with Embedding v2
• Fixed VOCR for PDF in certain edge cases
• Fixed AI Scraper returning invalid JSON for empty element_prompts
• Decrease latency for AI Scraper

​

DOCS

• Updated File Uploads documentation Handling Files
• Renamed embeddingV2 method to embedding_v2 for consistency

​

SDKs

• Updated the way multipart/form-data are handled for APIs involved with file uploads
• Renamed embeddingV2 method to embedding_v2 for consistency

​

Python

• Fixed blob inputs for Object Detection where previously features params was not respected
• Fixed blob inputs for VOCR where page range was not respected
• Established CI/CD pipeline for Python SDK
• Updated to pass x-jigsaw-no-request-log through headers. Security

​

JavaScript

• Fixed blob inputs for Object Detection where previously features params was not respected
• Fixed blob inputs for VOCR where page range was not respected

​

Features

• params: speaker_fingerprint a boolean value to switch on/off speaker identification.
• response: speaker_embeddings a list of embeddings for the speaker.

​

Fixes

• fixed Speech to Text (STT) webhook
• fixed file upload overwrites error

​

API

• added embedding v2 API

​

SDKs

• additional type safety for all APIs
• added embedding v2
• added deep research
• removed Text to Speech (TTS) from SDKs
• brought back prompt engine

​

DOCS

• changelog introduction
• examples for all SDKs supported
• Vercel Integration and Marketplace
• Handling Files documentation