Selected industry news from AIHOT. Follow advances in AI, products and technology.
AIHOT · Selected summaries with original sources · Nov 4, 2026, 9:00 AM
The GPT-4 API is now generally available; general access is also opened for the GPT-3.5 Turbo, DALL·E, and Whisper APIs; legacy models in the Completions API will be deprecated in early 2024.
OpenAI releases ChatGPT and Whisper APIs.
Hugging Face has launched the Open CoT Leaderboard, which evaluates large language models' ability to generate effective reasoning chains by calculating the difference in accuracy before and after applying chain-of-thought (Δ = accuracy with CoT – accuracy without CoT). It covers multiple choice-based reasoning tasks such as LogiQA and LSAT, employs two prompting strategies—Classic and Reflect—and six decoding configurations. Currently, 30 open-source models have been evaluated.
Hugging Face has released Jack of All Trades (JAT), a Transformer-based general-purpose agent model supporting sequential decision-making tasks such as video games, robot control, and navigation instruction understanding.
Hugging Face has released an open medical large language model leaderboard, evaluating accuracy based on MedQA, MedMCQA, PubMedQA, and medical subsets of MMLU (including clinical knowledge, anatomy, professional medicine, etc.).
Meta has released the Llama 3 open-source large language model, featuring two parameter sizes: 8B and 70B. Each size is available in base and instruct-tuned versions, totaling 4 main models; additionally, a safety model named Llama Guard 2, fine-tuned based on Llama 3 8B, was also released.
Mistral has released the Mixtral 8x22B open-source large language model, which adopts a sparse MoE architecture with 39B active parameters (141B total parameters) and supports a 64K token context window.
The Zama team has launched privacy-preserving inference services supporting Fully Homomorphic Encryption (FHE) on Hugging Face Endpoints. Users can one-click deploy pre-compiled models (such as concrete-ml-encrypted-decisiontree) and call the encrypted inference interface.
Gradio has introduced reload mode, allowing developers to load code changes in real time without restarting the server. This mode is implemented via custom reloading logic to achieve faster responses and selective reloading (such as skipping model loading). The article uses a document question-answering application based on the Hugging Face Inference API as an example to fully demonstrate the development process from UI adjustments to logic integration.
A Hugging Face blog post introduces a new programming LLM evaluation leaderboard based on LiveCodeBench. This benchmark was developed by researchers from UC Berkeley, MIT, and Cornell University. It collects problems chronologically from platforms such as LeetCode, AtCoder, and CodeForces. It supports four scenarios: code generation, self-repair, code execution, and test output prediction. All scenarios use the Pass@1 metric, and temporal evaluation prevents data contamination by leveraging problem publication dates.
Hugging Face has released Idefics2, an open-source multimodal vision-language model with 8B parameters under the Apache-2.0 license. It supports arbitrary interleaved image-text inputs and text generation, featuring enhanced OCR capabilities and native-resolution image processing. It outperforms models of similar size on benchmarks such as MMMU and is integrated into 🤗 Transformers. Accompanying resources include the open-source instruction-tuning dataset The Cauldron and a complete training data manifest.
OpenAI announced the establishment of its first office in Asia and released a custom GPT-4 model optimized for Japanese.
Hugging Face has launched a new "Deploy on Google Cloud" integration, supporting one-click deployment of open-source foundation models tagged with "text-generation-inference" from Hugging Face Hub (such as Zephyr Gemma) onto Vertex AI or GKE.
Google has launched the CodeGemma series of open-source code large language models, including a 2B base model (focused on code completion), a 7B base model (trained on mixed code and natural language), and a 7B instruction-tuned model (supporting code dialogue). All are based on Gemma pre-trained checkpoints and additionally trained on 500B tokens, with an 8K context window.
Hugging Face announced a partnership with cloud security company Wiz to integrate its Vulnerability Management and CSPM capabilities to enhance platform security. The Wiz research team discovered arbitrary code execution risks related to pickle in Hugging Face's sandboxed computing environment; all related issues have been fully remediated. Hugging Face is also promoting Safetensors as a secure alternative to pickle and continues to drive security best practices within the open-source AI community.
MotherDuck and Numbers Station launched the DuckDB-NSQL-7B model, fine-tuned based on Llama-2-7b, specifically for text-to-SQL tasks in DuckDB.
In a blog post, Hugging Face introduced how to use 🤗 Optimum Intel to perform post-training static quantization (INT8) on SetFit models, achieving up to a 7.8x improvement in inference throughput on Intel Xeon CPUs.
Hugging Face announced the launch of the Deploy on Cloudflare Workers AI integration, supporting serverless invocation of Llama 2 7B at Cloudflare edge data centers.
The Pollen Robotics team has released the open-source library pollen-vision, which provides a unified interface for zero-shot vision models such as OWL-VIT, Mobile SAM, and RAM. It supports 3D object localization (x, y, z coordinates) based on RGB images and text inputs, suitable for the Reachy humanoid robot's task of grasping unknown objects. The library can run in real time on consumer-grade GPUs (such as RTX 3070) and has been open-sourced on GitHub.
The Hugging Face blog introduces binarization (binary) and int8 scalar quantization methods for embedding vectors. Demonstrated in real-world retrieval scenarios on 41 million Wikipedia texts, quantization can achieve up to a 45x retrieval speedup.
Lighthouz AI and Hugging Face jointly launched Chatbot Guardrails Arena, a stress-testing platform for AI chatbot privacy protection capabilities; users can interact with two anonymous...
A Hugging Face blog post demonstrates how to use Optimum Intel and Intel OpenVINO to perform 4-bit weight quantization on the Microsoft Phi-2 model.
GaLore is a memory optimization technique for large language model training that significantly reduces optimizer state memory usage via low-rank gradient projection (reducing optimizer state memory during training by over 82.5%), enabling the training of Llama architecture 7B parameter models on consumer-grade GPUs such as the NVIDIA RTX 4090.
Hugging Face released Cosmopedia, an open-source synthetic dataset for LLM pretraining containing over 30 million files and 25 billion tokens, generated by Mixtral-8x7B-Instruct-v0.1.