Docling v2.140.0
IBM shipped Docling v2.140.0 (minor release) on GitHub.
Why it matters: New version with notable changes — read the release notes for breaking changes before upgrading.
All labs
Coverage & source status· 3/3 sources OK · checked Show
Official IBM Research RSS + ibm-granite on Hugging Face + GitHub.
- IBM Research BlogRSSsucceeded
- Hugging Face · ibm-graniteHugging Facesucceeded
- GitHub · docling-project/doclingGitHub Releasessucceeded
41 updates · page 1 of 3
Friday, October 16, 2026
Developer Thursday, October 15, 2026
Model granite-switch-4.1-8b
IBM published granite-switch-4.1-8b on Hugging Face for text generation under the Apache-2.0 license.
Why it matters: Teams that want a smaller Granite Switch model can use the 8B checkpoint under Apache-2.0.
Model granite-switch-4.1-30b
IBM published granite-switch-4.1-30b on Hugging Face for text generation under the Apache-2.0 license.
Why it matters: Developers can use this 30B Granite Switch 4.1 checkpoint under Apache-2.0.
Wednesday, October 14, 2026
Developer Docling v2.139.0
Docling v2.139.0 exposes max_concurrent_results, per-page confidence breakdowns, conversion progress bars, and picture reading in Numbers table cells, and can download whisper artifacts to run ASR.
Why it matters: Docling users on v2.139.0 can track conversion progress, read Numbers pictures, and run ASR after downloading whisper artifacts.
Tuesday, October 13, 2026
Feature Preparing chip designers for the nanostack era
IBM is releasing a process design kit for its 0.7-nanometer transistor architecture to prepare chip designers for the nanostack era.
Why it matters: Chip designers can use the kit to work with that 0.7-nanometer transistor architecture.
Research Scaling AI-driven chip design with open agent factories
Open models and llm-d enabled IBM Research to explore quantum error correction hardware at 11.5× lower inference cost than an all-proprietary configuration.
Why it matters: Chip-design teams can see how those open models and llm-d lowered inference cost for that hardware exploration.
Monday, October 12, 2026
Developer Docling v2.138.0
Docling v2.138.0 reads pictures placed on a Numbers sheet, emits DOCX textboxes once inside their table cell, and parses ordered lists and list item labels in JATS.
Why it matters: Document pipelines keep Numbers images, in-cell Word textboxes, and labeled JATS lists when converting files.
Model How distributed inference keeps large AI models moving
IBM explains distributed inference serving as the system that keeps large AI models moving, and says agentic workflows make it more necessary than ever.
Why it matters: Operators of large models can see why spreading inference across machines matters more for agent traffic.
Friday, October 9, 2026
Developer Docling v2.137.0
Docling v2.137.0 can add enrichments after the VLM pipeline and reads PP-OCRv6 languages through rapidocr’s supported-language list.
Why it matters: Document pipelines can enrich VLM output and pick up more OCR languages.
Developer Docling v2.136.0
Docling v2.136.0 recognizes Division and Subdivision headings, adds MLX repetition-penalty options, skips OCR when a vector shape crosses programmatic text, and resolves LaTeX input and images for string sources.
Why it matters: Conversions keep heading structure and LaTeX figures, including CMYK JPEGs.
Wednesday, October 7, 2026
Developer Docling v2.135.0
Docling v2.135.0 renders Apple Numbers chart images, reads iWork '09 chart kinds, and keeps inline formatting inside Markdown table cells.
Why it matters: Spreadsheet and Markdown conversions keep chart images and inline formatting in tables.
Feature What's new at IBM Quantum Q3 2026
IBM Quantum's Q3 2026 roundup collects the latest updates, releases, and resources from IBM Quantum and the Qiskit community.
Why it matters: Qiskit users can scan one Q3 2026 post for what shipped and which resources to read.
Tuesday, October 6, 2026
Developer Docling v2.134.0
Docling v2.134.0 adds Apple Numbers support, exposes logprobs on the VLM streaming path, and keeps text from dangling fragment hyperlinks in DOCX.
Why it matters: You can convert .numbers files and read token logprobs while a VLM stream is running.
API Qiskit beyond Python: Fortran, C++, and Julia
Qiskit's C API connects the SDK's Rust core to Fortran, C++, and Julia so you can build and run circuits natively without Python.
Why it matters: HPC and scientific computing teams can call Qiskit from Fortran, C++, or Julia.
Monday, October 5, 2026
Research Directed execution: same performance, more control
IBM describes directed execution so you can see, customize, and extend how circuits use error mitigation and correction on IBM quantum hardware, with the same performance and more control.
Why it matters: Quantum users can choose which error mitigation and correction methods a circuit applies.