Introducing GPT-6 Sol and Luna
OpenAI has released two new models, GPT-6 Sol and GPT-6 Luna. They offer different trade-offs between capability and cost for everyday work.
Week of September 21, 2026
September 21, 2026 – September 27, 2026 · 34 storiesOpenAI released two new models, GPT-6 Sol and GPT-6 Luna, which trade off capability against cost, and improved prompt caching for GPT-6. CEO Sam Altman addressed the UN Security Council on AI safety and international cooperation, as OpenAI also proposed a path toward shared international AI standards. NVIDIA launched DSX Ready to qualify power and cooling equipment for AI data centers and released Isaac ROS 5.0 for robotics developers. Google DeepMind added text-to-speech and a Live Avatar feature to Gemini 3.8, and Hugging Face shipped version 1 of its tokenizers library.
Model releases
Gemini 3.8 text-to-speech says hello
Google DeepMind has announced text-to-speech capabilities for Gemini 3.8.
Introducing Gemini 3.8 Live with Live Avatar
Google DeepMind has introduced a Live Avatar feature for Gemini 3.8 Live.
Open source
tokenizers v1: encode, decode and scaling, measured
Hugging Face has released version 1 of its tokenizers library, with measurements of encoding, decoding and scaling performance.
Transformers now runs llama.cpp quants
Hugging Face’s Transformers library can now run models quantized in the llama.cpp format.
Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community
Jun Kim, who created and maintains oMLX, is joining Hugging Face. His role will focus on supporting the community around MLX, Apple’s machine learning framework.
NVIDIA Isaac ROS 5.0 Advances Agentic, Open Source Robotics Development
NVIDIA has released Isaac ROS 5.0, a set of GPU-accelerated packages built on the open ROS robotics framework. It targets developers building robots that perceive, reason and act in changing environments.
Agents & harnesses
V7 cuts costs 78% while boosting accuracy with GPT-5.6 Luna
V7 uses GPT-5.6 to turn scattered company files into context that agents can draw on for complex tasks with linked sources. The company reports a 78% cost reduction alongside better accuracy.
Parallel cut research time and cost in half with GPT‑6 Astra
With GPT-6 Astra, Parallel’s agents researched and synthesized labor-market data in half the time and at half the cost of earlier models, according to OpenAI.
Ringg’s AI agents resolve up to 65% of customer calls with OpenAI
Ringg uses GPT-5.6 to run multilingual customer-service agents across voice, chat, WhatsApp and the web. The company says they resolve up to 65% of calls at 90% lower cost than with GPT-4.1.
Proaction boosts sales 60% and saves 75+ hours with Codex
Fleet-management company Proaction uses Codex, GPT-Live-1 and GPT-6 Astra to build, run and sell its software faster. It reports a 60% increase in sales and more than 75 hours saved.
Research
Advisory Group on Mathematics and Artificial Intelligence
OpenAI is working with an independent advisory group of mathematicians. The group will help guide how new AI-generated mathematical results are reviewed and communicated.
How UK AISI and EvalEval Are Making Benchmark Results Reproducible
A Hugging Face post covers work by the UK AI Security Institute and the EvalEval initiative to make AI benchmark results reproducible.
Products & apps
Expanding OpenAI Academy with new learning paths
OpenAI Academy is adding learning paths for employees, developers, managers, teachers and students. They are meant to help people build and demonstrate practical AI skills.
Higgsfield AI ships new video features in a day with GPT-6 Astra
Higgsfield AI uses GPT-6 Astra to simplify video ad production for small businesses. The company says it can now release new creative tools much faster.
Better prompt caching for GPT-6
OpenAI has improved prompt caching for GPT-6 with higher hit rates, new diagnostics, explicit breakpoints and extra controls. The changes are meant to lower latency and cost for developers.
Airbnb widens access to GPT-6 Astra and OpenAI frontier models
Airbnb is giving more of its engineering teams access to GPT-6 Astra and other OpenAI models. The goal is to help them fix bugs, design systems and ship faster.
ChatGPT Ads expands to Southeast Asia and Taiwan
OpenAI is extending ChatGPT Ads to Southeast Asia and Taiwan. Eligible businesses can now advertise to users in more than 60 countries.
How invideo improves color grading 3x with GPT‑6 Astra
Video platform invideo uses GPT-6 Astra to plan edits more precisely. The company reports a threefold improvement in color correction and grading and says it produced 50 custom effects in a single day.
Harvey turns legal context into stronger drafts with GPT-6 Astra
Legal AI company Harvey says GPT-6 Astra generates better-structured legal documents that take more context into account. This leaves lawyers more time for strategy.
Two years of OpenAI Academy
OpenAI marks two years of its OpenAI Academy training program and says it will bring AI skills to more communities.
Google Beam expands with new regions, partners, and customers
Google says its Beam video communication platform is expanding to new regions and adding partners and customers.
Wayfair boosts catalog accuracy and support speed with OpenAI
Online retailer Wayfair uses OpenAI models to automate support ticket triage. It also uses them to improve millions of product attributes in its catalog.
Lowe’s answers millions of home improvement questions with OpenAI
Home improvement chain Lowe’s uses OpenAI models to power its Mylow and Mylow Companion assistants. They have answered more than 25 million questions from customers and store staff.
Chips & compute
NVIDIA Launches DSX Ready to Qualify Power and Cooling Products for AI Factories
NVIDIA has introduced DSX Ready, a program that qualifies power and cooling products for AI data centers. It is meant to help builders choose equipment that fits the overall design of their facility amid power, water and grid constraints.
At AI Day Singapore, NVIDIA and Partners Showcase AI Advancements Across Southeast Asia
NVIDIA held its AI Day in Singapore on Sept. 22-23 with training sessions and expert talks on AI and high-performance computing. The company and its partners showcased AI projects from across Southeast Asia.
Business & funding
Grab and OpenAI bring practical AI skills to Southeast Asia
OpenAI and Grab have launched GO Forward with AI, a regional program to teach practical AI skills. It targets 30,000 Grab partners across Southeast Asia.
Policy & regulation
Building standards for the next phase of AI
OpenAI proposes a route toward shared international AI standards. It calls for coordinated evaluation, reporting and governance to make AI safer.
Sam Altman’s remarks at the United Nations Security Council
OpenAI CEO Sam Altman addressed the UN Security Council. He spoke about AI safety, keeping humans in control and the need for international cooperation.
Safety & security
Why Deploying Physical AI at Scale Demands Safety at Every Layer
NVIDIA argues that robots and autonomous vehicles need safety built into every layer as they move from research into shared spaces. It cites analyst forecasts of tens of millions of autonomous vehicles and industrial robots deployed over the next decade.
Priorities and principles for effective third party assessments
OpenAI sets out what it considers necessary for independent safety assessments of frontier models and their safeguards. It stresses rigor, security and independence.
Introducing MentalHealthBench
OpenAI has released MentalHealthBench, a benchmark designed with expert input. It assesses whether AI responses are helpful and safe in realistic conversations about mental health.
OpenAI extends cyber access to Ukraine for civilian defense
OpenAI is giving Ukraine’s government access to its Daybreak program. The aim is to support the cyber defense of civilian infrastructure.
Advancing Private AI Compute with secure, server-side memory
Google is adding private, server-side memory to Private AI Compute, its system for personal AI. The feature is designed to keep users’ data protected.