Product Launches
Google's PaperOrchestra AI Paper-Writing Framework Adds 1,529 GitHub Stars in a Day
By News Desk ·
Key takeaways
- PaperOrchestra gained 1,529 GitHub stars in 24 hours on November 18, 2026, reaching 3,692 stars and 469 forks.
- The Python framework uses specialized agents for outline generation, literature review, section writing, content refinement and visuals.
- PaperOrchestra produces LaTeX manuscripts formatted to target conference templates and ranks #15 on GitHub's trending list.
Google-research's open-source PaperOrchestra project gained 1,529 GitHub stars in 24 hours on November 18, 2026, reaching 3,692 stars and 469 forks. Blockchain News reported the surge, noting the repository ranks #15 on GitHub's trending list.
How PaperOrchestra Writes a Paper
Google AI's PaperOrchestra is a multi-agent framework written primarily in Python that automates the path from raw research materials to a submission-ready manuscript. According to the project's README, it runs as an orchestrated team of specialized agents, each handling a distinct task.
One agent generates the outline that structures the paper. A literature review agent searches, prioritizes and integrates relevant academic references. Section-writing agents draft individual parts of the manuscript, and a content refinement agent polishes the drafted text. A visuals agent creates plots and conceptual diagrams for inclusion in the paper.
The framework also automates production of LaTeX manuscripts that follow target conference templates, which the project describes as useful for researchers meeting specific formatting requirements. Setup requires creating a Conda environment with Python 3.11 and installing dependencies from a requirements file. Users configure a .env file with API keys for services such as Semantic Scholar, OpenAI or Google's Gemini models, depending on the functionality they want.
The project's usual activity is approximately eight stars per week, against the 1,529 stars added in a single day. The repository is available through the Google Research site.
Google AI's Open-Weight and Agent Record
Google AI's recent company records, according to akta.pro company records, include the June 2026 launch of Gemma 4 12B, a unified encoder-free multimodal open-weight model under Apache 2.0 aimed at 16GB-VRAM laptops, with Multi-Token Prediction drafters delivering up to 3x faster inference. The same records show the Gemma line has crossed 150 million downloads.
Earlier Google Research Launches
Google Research unveiled Language Explorer on November 9, 2026, an interactive platform covering about 7,000 languages and 159 scripts, Unite.AI reported. On November 3, 2026, Google AI & Infrastructure announced Physical-Engineering's Last Exam, a benchmark with 29 tasks across four engineering disciplines; all frontier models scored below 50% on a private test set, with the highest reaching 48.3%, according to the Google Cloud Blog.
Google Cloud AI Research open-sourced Harness² on October 18, 2026, a recursive agent harnessing framework that refines an agent's harness without changing model weights, MarkTechPost reported. A single refinement step improved GLM-5's WorkBuddy-Bench score by 11.5 points, from 64.2 to 75.7.