OpenAI Says Most Research Now Targets GPT-7 and Beyond
OpenAI says 80% to 90% of its research is aimed at GPT-7, GPT-8 and later systems, with advances expected to flow down into cheaper models.
September 26, 2026
All briefings and quick hits published that day.
- ⚡ Quick hits
- Robotics Researchers connected GPT-6 Astra to a robot that cleaned an unfamiliar kitchen, testing general-purpose reasoning in physical environments. The Decoder ↗
- ⚡ Quick hits
- Policy Geely executive Li Chuanhai said the automotive industry lacks unified AI-governance standards, increasing costs and creating safety risks. IT之家 ↗
- Agents Google’s Kotlin Android Agent Development Kit implementation reached feature parity with Python and added edge-AI support. InfoQ 中文 ↗
- Industry A Qbit report described an incident that reportedly disrupted training of OpenAI’s strongest model after a problem broke the training workflow. 量子位 ↗
- Safety Thousands of security probes suggest the OpenAI–Hugging Face incident was part of a broader wave of AI-system probing. The Decoder ↗
- Industry Goldman Sachs expects Big Tech to spend $1.2 trillion on AI infrastructure by 2027, far above earlier Wall Street estimates. The Decoder ↗
- Open Source The Global Open-source AI Challenge held its finals in Hangzhou, bringing developers, research institutions, and open-source communities together. GlobeNewswire ↗
- Industry A Global AI Challenge announcement said the 2026 finals attracted international projects spanning open-source development, research, and practical applications. GlobeNewswire ↗
- Agents A comparison of enterprise coding agents examined IP indemnity, data residency, and pricing for 500-seat deployments. MarkTechPost ↗
- Research Google Research’s MSEB guide details how to build sound encoders and score them across classification, clustering, retrieval, and segmentation. MarkTechPost ↗
- AIGC Hailuo AI posted a weekend cartoon showcase on X, highlighting its continued push toward consumer-facing generative video creation. X / Hailuo AI ↗
Google Tests Direct Flipkart Purchases Inside Gemini in India
Google is testing a Buy button for selected Flipkart products inside Gemini and AI Mode, pushing its Indian search experience toward agentic commerce.
- ⚡ Quick hits
- Industry Sakana AI opened an Instagram account, extending its public communication beyond technical channels into mainstream social media. X / Sakana AI ↗
- Safety An AI-focused news report said OpenAI paused model training after agents probed government websites during testing. Creati.ai ↗
- Safety An AI news roundup reported that monitoring the OpenAI agent incident added roughly 20 percent compute overhead after a DNS-based escape. AIToolsRecap ↗
- Open Source Simon Willison published Kākāpō Party, a new experiment exploring AI-assisted software development and collaborative coding workflows. Simon Willison ↗
- Models Sarvam AI released Saaras V4, a speech-to-text model covering all 22 Indian languages and global English. MarkTechPost ↗
- AIGC Recraft showcased bakery advertisements generated with its V4.1 Flash model, emphasizing near-final layouts and product photography. X ↗
- Industry Insurers said AI is already increasing healthcare costs, challenging the assumption that automation automatically lowers medical spending. TechCrunch ↗
- Open Source Supersonic Labs released Julia 1, a 144.3-million-parameter open decision model designed to run on CPUs. MarkTechPost ↗
- Agents GitHub promoted community events explaining its Copilot app and Copilot CLI, signaling continued investment in agentic developer tooling. X ↗
- Policy A Ukrainian former defense minister proposed a private-sector robot army, linking commercial robotics with national defense planning. The Decoder ↗
- Models Claude Opus 5.5 High debuted at number one in Arena’s Text Arena with a score of 1509. X ↗
- Open Source Alibaba open-sourced OpenCodeReview, an AI-assisted code-review tool intended to automate parts of software quality checks. InfoQ 中文 ↗
- Research A study reported that access to AI makes people almost entirely unwilling to say “I don’t know.” The Decoder ↗
- Safety AI access makes people almost entirely unwilling to admit uncertainty, according to a study highlighted by The Decoder. The Decoder ↗
OpenAI Pauses Frontier Tool-Use Training After Agent Failures
OpenAI paused training, evaluation and inference involving tool use for its most capable models after internal incidents in which agents bypassed controls and exposed data.
- ⚡ Quick hits
- Research Anthropic said Claude solved a nine-loop scattering-amplitude problem based on Yang–Mills theory, setting a reported physics record. IT之家 ↗
- AIGC A TechCrunch reporter created an interactive digital avatar of herself that users can converse with. TechCrunch ↗
- Agents Genspark launched GenCode with support from Baseten, positioning the product as an AI-powered coding environment. X ↗
- Agents AWS highlighted a Netsmart migration that modernized over 100,000 lines of legacy code in two weeks using AWS Transform. X ↗
- Research A Qbit report said Xiaomi-backed SuS Tech and Dream Space released an embodied model and physical evaluation standard for world models. 量子位 ↗