AI News — HefuAI Platform

• Zhipu GLM 5.3 Model Overview: 4K Context Window and Cost-Effective Chat Capabilities — GLM 5.3 is a chat model developed by Zhipu, featuring a 4,096-token context window with official pricing of ¥8 for input and ¥28 for output per million tokens. 2026-09-30
• Making AI an asset, not an expense — As AI transitions from experimentation to production, the traditional consumption-based pricing model can lead to unpredictable monthly expenses for steady, bus 2026-09-29
• Introducing GPT-6.1 Sol: Near-Astra Intelligence at One-Fifth the API Cost — OpenAI has introduced GPT-6.1 Sol, a new artificial intelligence model designed to deliver near-Astra intelligence for coding, computer use, and professional wo 2026-09-29
• DevDay 2026 Recap — OpenAI's DevDay 2026 featured more than 20 major announcements across multiple areas of the company's ecosystem. Key highlights included the release of GPT-6 As 2026-09-29
• Zhipu GLM 5.3 Flash Model Review: Cost-Effective Chat Solution — Zhipu's GLM 5.3 Flash model is designed for efficient chat scenarios. Featuring a 4,096 tokens context window, it meets daily text interaction needs. The pricin 2026-09-29
• Zhipu GLM-5.2 Model Analysis: 1M Context and Cost-Effective Long-Task Engineering — Zhipu's GLM-5.2 is a flagship chat model designed for long-task execution. It supports a genuinely usable context window of 1,000,000 tokens, maintaining high s 2026-09-29
• Introducing dots: OpenAI's Proactive Assistant for Complex Projects and Everyday Tasks — OpenAI has introduced "dots," a new proactive assistant designed to keep working across both complex projects and everyday tasks. The core feature of dots is it 2026-09-29
• When can we say AI made a scientific discovery? — Anthropic recently announced that its AI-powered biology lab made its first discovery, with Claude agents identifying a repeating genetic pattern reminiscent of 2026-09-28
• The Pentagon wants $30 million to build an AI-powered lie detector — The US Department of Defense plans to invest $30.3 million over five years to develop an AI-powered lie detector called Polygraph+. Run by the Defense Counterin 2026-09-28
• OpenAI extends cyber access to Ukraine for civilian defense — OpenAI has announced the extension of access to its Daybreak program to the Government of Ukraine, specifically aimed at supporting the cyber defense of the nat 2026-09-28
• Holo4: powering generalist computer-use agents — Hcompany has released Holo4, a new series of agentic models available in 27B dense and 35B-A3B Mixture of Experts sizes, alongside Holotron4 Nano. Holo4 interac 2026-09-28
• DeepSeek V4 Flash Model Analysis: Million-Token Context and High Cost-Performance — DeepSeek V4 Flash is a highly cost-effective chat model by DeepSeek, balancing response speed and reasoning quality. It supports a 1,000,000-token context windo 2026-09-28
• Who’s liable when AI agents go rogue? — A recent wave of cyberattacks by rogue AI agents from OpenAI, Anthropic, and Google has raised urgent questions about corporate liability. Current state AI tran 2026-09-28
• OpenAI GPT-6 Sol Model Focus: 1.05M Context & Cost-Effective Coding — OpenAI's GPT-6 Sol (model ID: gpt-6-sol) is a high-performance model designed for programming and daily tasks, featuring a massive context window of 1,050,000 t 2026-09-28
• Anthropic Claude Fable 5 Model Focus: Advanced Reasoning and Autonomous Agent Capabilities — Claude Fable 5 is an Anthropic Mythos-level flagship model featuring top-tier reasoning and long-term autonomous agent capabilities. With a context window of 13 2026-09-28
• Claude Opus 5.5: Anthropic's Advanced Model for Long-Horizon Agentic Coding and Knowledge Work — Claude Opus 5.5 is Anthropic's advanced model for long-horizon agentic coding and knowledge work. It features always-on adaptive thinking and strong tool-callin 2026-09-28
• DeepSeek V4.1 Flash Model Overview: A Cost-Effective Solution for Million-Token Context and High-Concurrency Chat — DeepSeek V4.1 Flash is a highly cost-effective chat model developed by DeepSeek, specifically designed for high-concurrency general conversations and long-text 2026-09-28
• OpenAI GPT-6 Luna Model Review: Million-Token Context and High Cost-Efficiency for Daily Tasks — GPT-6 Luna is a fast and cost-effective model introduced by OpenAI for daily tasks. Featuring a massive context window of 1,050,000 tokens and supporting the ch 2026-09-28
• The Lenfest Institute Grows Landmark Program with Expanded OpenAI Support — OpenAI is significantly expanding its support for the Lenfest AI Collaborative and Fellowship Program, a landmark initiative driven by the Lenfest Institute. Th 2026-09-28
• Basis completes a tax workbook 2x faster with GPT-6 Astra — Basis recently tested the new GPT-6 Astra model on a complex tax task, demonstrating significant performance improvements over its predecessor. The new model su 2026-09-28