InfoQ Homepage Large language models Content on InfoQ
-
Hugging Face Releases FinePDFs: a 3-Trillion-Token Dataset Built from PDFs
Hugging Face has unveiled FinePDFs, the largest publicly available corpus built entirely from PDFs. The dataset spans 475 million documents in 1,733 languages, totaling roughly 3 trillion tokens. At 3.65 terabytes in size, FinePDFs introduces a new dimension to open training datasets by tapping into a resource long considered too complex and expensive to process.
-
Cloudflare Introduces Automated Scoring for Shadow AI Risk Assessment
During AI Week 2025, Cloudflare announced Application Confidence Scores, an automated assessment system that is designed to help organizations evaluate the safety and security of third-party AI applications at scale.
-
Vercel Introduces AI Gateway for Multi-Model Integration
Vercel has rolled out the AI Gateway for production workloads. The service provides a single API endpoint for accessing a wide range of large language and generative models, aiming to simplify integration and management for developers.
-
Google DeepMind Launches EmbeddingGemma, an Open Model for On-Device Embeddings
Google DeepMind has introduced EmbeddingGemma, a 308M parameter open embedding model designed to run efficiently on-device. The model aims to make applications like retrieval-augmented generation (RAG), semantic search, and text classification accessible without the need for a server or internet connection.
-
xAI Releases Grok Code Fast 1, a New Model for Agentic Coding
xAI introduced grok-code-fast-1, a model developed specifically for agentic coding workflows.
-
Google Launches Gemini 2.5 Flash Image with Advanced Editing and Consistency Features
Google released Gemini 2.5 Flash Image (nicknamed nano-banana), its newest image generation and editing model. The system introduces several upgrades over earlier Flash models, including character consistency across prompts, multi-image fusion, precise prompt-based editing, and integration of world knowledge for semantic understanding.
-
LinkedIn Re-Architects Edge-Building System to Support Diverse Inference Workflows
LinkedIn has detailed its re-architected edge-building system, an evolution designed to support diverse inference workflows for delivering fresher and more personalized recommendations to members worldwide. The new architecture addresses growing demands for real-time scalability, cost efficiency, and flexibility across its global platform.
-
DeepSeek Releases v3.1 Model with Hybrid Reasoning Architecture
DeepSeek has released version V3.1 of its large language model, introducing a hybrid architecture that combines thinking and non-thinking modes in a single system.
-
Anthropic’s Claude Opus 4.1 Improves Refactoring and Safety, Scores 74.5% SWE-bench Verified
Anthropic has launched Claude Opus 4.1, an update that strengthens coding reliability in multi-file projects and improves reasoning across long interactions. The model also raised its SWE-bench Verified score to 74.5%, up from 72.5%. Building on Opus 4, the new version strengthens Claude’s ability to act as a coding assistant, particularly in multi-file contexts.
-
Qwen Team Open Sources State-of-the-Art Image Model Qwen-Image
Qwen Team recently open sourced Qwen-Image, an image foundation model. Qwen-Image supports text-to-image (T2I) generation and text-image-to-image (TI2I) editing tasks, and outperforms other models on a variety of benchmarks.
-
Claude Sonnet 4 Expands to 1 Million Token Context Window
Anthropic has upgraded Claude Sonnet 4 to support a context length of up to 1 million tokens, a fivefold increase over its previous limit. The feature, now in public beta, is accessible through the Anthropic API and Amazon Bedrock, with Google Cloud’s Vertex AI support expected soon.
-
DeepMind Launches Genie 3, a Text-to-3D Interactive World Model
DeepMind has introduced Genie 3, the latest version of its “world model” framework for generating interactive 3D environments directly from text prompts.
-
Unsloth Tutorials Aim to Make it Easier to Compare and Fine-tune LLMs
In a recent Reddit post, Unsloth published comprehensive tutorials of all of the open models they support. The tutorials can be used to compare the models’ strengths and weaknesses, as well as their performance benchmarks.
-
Anthropic Investigates How Large Language Models Develop a Character
Recent research by Anthropic engineers explores identifiable patterns of activity that seems to give rise to an emerging personality. These traits, known as persona vectors, help explain how a model's personality shifts over its lifecycle and lay the groundwork for better controlling those changes.
-
Google Launched LangExtract, a Python Library for Structured Data Extraction from Unstructured Text
Google has introduced LangExtract, an open-source Python library designed to help developers extract structured information from unstructured text using large language models such as the Gemini models.