AI News Summary 2026-08-26
.. lang: en
AI News Summary 2026-08-26
GAFAM and Major AI Companies
OpenAI Integrates Full-Stack Strategy, In-House Chip, and Enterprise Management
OpenAI News published “The full stack behind abundant intelligence” on August 25, highlighting the first results from Jalapeño and the Admin plugin for ChatGPT Work and Codex. The key takeaway is that OpenAI is presenting AI as a complete stack: data centers/chips, state-of-the-art models, a development platform, consumer/enterprise products, and AI-native devices. Jalapeño is its first in-house inference chip, offering improvements in throughput per kW and latency on InferenceX.
Anthropic Funds Better Well-Being Assessments
Anthropic News announced on August 25 “Funding better evaluations of AI’s impact on well-being,” an initiative focused on better measuring how AI systems affect well-being before scaling up their use.
Influencers and Tech Blogs
Simon Willison Summarizes the Debate on AI and Trustworthy Software at Scale
Simon Willison’s Weblog published a post on August 26 quoting Paul Dix on AI writing/refining 1M LOC to turn it into reliable software used by millions of machines. In the same recent post, Willison also covered EVE Online’s move to Python 3 and the compatibility of Anthropic’s plugin for llm with anthropic v1.0.0.
OpenRouter Brings Model Selection to the Editor
OpenRouter Announcements posted “How to Choose the Best AI Model (Live, in Your Editor)” on August 25, a sign that model comparison is beginning to be integrated more directly into the development workflow.
Generative Imaging
Stability AI Receives Support from the Entertainment Industry
Stability AI News announced on August 25, “The Entertainment Industry’s Biggest Names Back Stability AI in Latest Funding Round,” with a total Series B funding of 232M and backers from the entertainment industry.
OpenRouter Expands Image Benchmarks and Video Generation Guide
OpenRouter Announcements posted “OpenRouter Video Generation API: A Code-First Guide” on August 25 and “Image Benchmarks: See the Capabilities of Every Model," combining a video integration guide with a comparative benchmark for image model capabilities.
Chatbots and Agents
OpenAI Adds Administration Features for ChatGPT Work and Codex
OpenAI News posted “Introducing the Admin plugin for ChatGPT Work and Codex” on August 25. The plugin allows admins to analyze workspace information, update settings, and act on requests in ChatGPT Work/Codex.
Hermes Agent v0.20.5 Follows the Open-Source Agent Release Cycle
The GitHub API lists v2026.8.19 / Hermes Agent v0.20.5, released on August 21, as the latest release of NousResearch/hermes-agent.
On-Premise AI and Serving
Perplexity and Ollama Bring Local AI to User Workflows
Perplexity Blog announced on August 25 Portable Computer for local-first AI: it runs entirely on-device with NVIDIA, keeps data private locally, and uses the cloud on demand. Ollama Blog announced on August 25 support for Claude Desktop with Ollama, allowing local open-source models to work with Claude Desktop as a third-party gateway provider.
vLLM, llama.cpp, and Granite Speech Strengthen Open and Edge Serving
The GitHub API lists vllm-project/vllm v0.28.0, released on August 26, and ggml-org/llama.cpp v0.3.0, released on August 25. Hugging Face Blog published “Extremely Fast and Accurate Transcription with Granite Speech 5.0 Turbo CTC” on August 25, featuring compact 470M ASR models in English, 12,600+ RTFx on H200, public OpenASR results, and compatibility with edge devices.