AI News: August 27, 2026

1. IBM Releases Open-Weight Granite 4.2 Family With Agentic Capabilities

IBM. IBM released the Granite 4.2 family under an Apache 2.0 license, spanning 3B, 8B, and 30B parameter models trained from scratch on roughly 15 trillion tokens with context windows up to 512,000 tokens. The larger 8B and 30B models add built-in agentic features including tool use, code execution, and sandboxed web search, plus a toggle between thinking and non-thinking modes and OpenAI-format tool calling. The models are available on Hugging Face, Ollama, and GitHub, alongside a new Granite Speech 5.0 Turbo CTC model that IBM says transcribes three hours of audio in one second. Source

2. Alibaba Launches Qwen3.8-Flash-Next Aimed at Cost Efficiency

Alibaba. Alibaba released Qwen3.8-Flash-Next, a multimodal mixture-of-experts model with 125 billion total parameters and only 6 billion activated per token, including a 51-billion-parameter N-gram embedding layer that runs in system RAM rather than GPU memory. The company says it outperforms Qwen3.7-Plus while costing about one-ninth as much to train, posting scores such as 91.9 on LiveCodeBench v6 and 91.7 on GPQA Diamond, with native context of 262,144 tokens scaling to one million via YaRN. Weights and a technical report are on GitHub and Hugging Face, with API pricing of $0.16 per million input tokens and $0.47 per million output tokens. Source

3. Z.ai Revealed as the Lab Behind the Anonymous Ox Alpha Model

Z.ai. Chinese research lab Z.ai confirmed it built Ox Alpha, a reasoning model that launched anonymously on OpenRouter before topping several coding and agentic benchmarks against frontier models from US labs. Z.ai, the maker of the GLM series including the recent GLM-5.3, described Ox Alpha as designed for coding, sustained agentic work, and production workloads, and said it planned to release the open weights. The launch adds to a wave of low-cost, capable Chinese open-weight models pressuring more expensive Western offerings. Source

4. Moonshot AI in Talks to Host Its Models on US Cloud Platforms

Moonshot AI. Chinese AI firm Moonshot AI, the developer of the Kimi models, is negotiating deals to distribute its models through Microsoft, Amazon, and Google cloud platforms, according to The Decoder. The arrangement would give a leading Chinese lab unprecedented access to US enterprise customers via major hyperscalers. It reflects growing demand for Chinese open-weight models even as export and hosting arrangements remain politically sensitive. Source

5. AI Assistant Startup Instinct Raises $350M at a $2.5B Valuation

Instinct. Instinct, built by Spear Street Technology and led by 23-year-old founder Noah Shinn, has raised $350 million in total including a $250 million Series B, reaching a $2.5 billion valuation with Index Ventures and Benchmark leading. The one-year-old startup, still in private beta, offers an AI assistant that connects to a user’s apps and devices to handle tasks like planning trips, buying tickets, and canceling subscriptions. Its rapid ascent has drawn scrutiny over the broad permissions and data-use terms the assistant requires. Source

6. Bill Gates Calls for a Robot Tax and ‘Human Reserved’ Jobs

Bill Gates. Bill Gates argued that AI is more dangerous than much of the tech industry will admit and proposed policy responses including a robot or token tax and categories of “Human Reserved” jobs to cushion labor disruption. He warned about risks ranging from mass unemployment to bioterrorism and called for nuclear-style oversight of advanced systems. The comments add a prominent voice to debates over how governments should manage AI’s economic and safety impacts. Source

7. Pro-Kremlin Deepfakes Impersonate Ukrainian Lawmakers

Ukraine. Researchers documented a Russian-aligned influence operation that used AI-generated deepfakes to put surrender rhetoric into the mouths of Ukrainian lawmakers, according to The Decoder. The fabricated clips drew around 130,000 views and continued to erode public trust even after being debunked. The campaign illustrates how synthetic media is being weaponized to target political figures in active conflict zones. Source

8. Ex-Meta Scientists Launch Perceptron for Factory-Floor Visual AI

Perceptron. Perceptron, founded by former Meta scientists, is building an AI model that gives machines visual intelligence and the ability to navigate physical environments, targeting industrial and factory-floor deployments. The startup positions its system as a foundation for perception in robotics and automation hardware. It joins a growing group of teams applying vision models to real-world physical operations. Source

9. Radar Makes 130,000 Podcasts Searchable for AI Agents

Particle. Particle launched Radar, a platform that transcribes more than 130,000 podcasts and exposes them through an API so AI agents and applications can search and use the audio content. The service aims to turn a large, hard-to-index corpus of spoken media into structured data for downstream tools. It targets developers building retrieval and agent systems that need access to podcast information. Source

10. Runable Raises $21M to Push AI Agents Into Running Businesses

Runable. Runable raised $21 million to develop AI agents that go beyond building businesses to operating and growing them over time. The company said 60 to 70 percent of its token usage comes from paying customers, a signal of real usage rather than free experimentation. The round reflects continued investor interest in agentic systems aimed at end-to-end business workflows. Source

11. Legato Emerges From Stealth With $12M and AI Hearing Glasses

Legato. Hearing-technology startup Legato came out of stealth with $12 million in funding and a preview of AI-enabled hearing glasses that integrate assistive audio into eyewear frames. The product aims to combine hearing support with everyday glasses rather than traditional standalone hearing aids. It represents an effort to apply on-device AI to accessibility and consumer wearables. Source

12. Arga Labs Raises $10M to Train Enterprise AI Agents

Arga Labs. Arga Labs raised $10 million led by General Catalyst to build tools for training enterprise AI agents. The company focuses on helping businesses develop and deploy agents tuned to their internal workflows and data. The funding targets a gap between general-purpose models and reliable, task-specific enterprise automation. Source

13. India’s Ringg Raises $10M From Peak XV for Voice AI

Ringg. Indian startup Ringg raised $10 million in a Series A extension backed by Peak XV Partners to expand its voice AI technology beyond phone calls into broader conversational applications. The company is building voice agents aimed at automating customer interactions across channels. The round highlights sustained investor appetite for voice-first AI in emerging markets. Source