[DigitalToday reporter Chi-gyu Hwang] Jev, a new AI model released by TypeSafe AI, a startup founded by a former OpenAI employee, has rapidly emerged as a buzzword in the AI arena as it positions itself as different from existing large language models (LLMs). Influential figures appear to be offering a brief comment on Jev in passing. A report also said TypeSafe AI is pushing to raise $1 billion in investment.

Experts say Jev is fundamentally different from LLMs in structure. It does not generate text. Rather than writing and holding conversations like ChatGPT, Jev is focused on making decisions in place of software. It is raising expectations that it has the potential to significantly improve the efficiency of AI agents.

• A new AI model released by a ChatGPT co-developer draws attention in the developer world • Why an AI model that judges and decides, rather than converses, is surprising • AI automation and work assistance cannot coexist... why? • An ambitious scenario driven by the unbundling of massive intelligence • Hot AI startup TypeSafe AI seeks $1 billion in funding... a $10 billion valuation?

OpenAI and Anthropic each unveiled new AI models with lower costs. It is part of an effort to respond to customer demand to reduce AI spending and to keep cheaper open-weight models in check.

• OpenAI-Anthropic unveil new AI models with lower prices... a warning shot to the open-weight camp • OpenAI, Anthropic, xAI roll out new AI models with lower prices... increased AI usage expected

Interest in personal AI agents appears to be growing further as Meta's Muse has posted an impressive early report card. OpenAI is also expected to enter the race soon. OpenAI earlier this year hired developers of OpenClaw, an open-source AI agent tool that gained popularity among AI power users, and put them in charge of developing a next-generation personal agent.

• Meta AI agent Muse surges... tops the App Store, beating ChatGPT • How Meta turned ordinary chatbot Muse into an AI talker • Meta AI agent Muse shakes up a subscription business model built on forgetfulness and hassle • "Competition in personal AI agents is a battle for personal data... it must dig into users' daily lives"

The article also summarizes moves by domestic and overseas companies around AI.

OpenAI added voice-based agent features to the ChatGPT mobile app. Users can handle tasks such as writing documents, drafting emails and summarising Slack messages on their smartphones using voice.

• OpenAI adds voice-based agent features to ChatGPT mobile app

Microsoft unveiled a new Copilot app that combines chatting, coding and AI agent features into one. Microsoft also changed the name of its AI personal assistant "Scout," introduced at this year's Build conference, to "Autopilot."

• Microsoft unveils Copilot "super app"... chat, coding and agents in one

Amazon Web Services (AWS) launched Strands Harness, an open-source agent designed to help developers build and deploy AI agents in any environment.

• AWS launches open-source AI agent Strands Harness for multicloud

Google is expected to release its next-generation AI model, "Gemini 4," earlier than its original target of the end of this year. YouTube unveiled an interactive editing tool in which AI refines long-form and short-form videos when users give editing instructions in natural language.

• Google, left behind by rivals, prepares a counterattack with Gemini 4... release imminent • YouTube unveils interactive tool for editing videos in natural language

Akamai Technologies signed a deal to supply Anthropic with $11.6 billion worth of computing capacity over 7 years. Databricks acquired early-stage spreadsheet startup Row Zero. Databricks plans to integrate spreadsheets as a work interface linking AI agents and business intelligence (BI).

• Akamai signs $11.6 billion computing supply deal with Anthropic • Databricks in another M&A deal... this time spreadsheet startup Row Zero

Liang Wenfeng (량원펑), chief executive of DeepSeek, identified training AI models with domestically produced chips as the top priority for investors. Liang said at a closed investor meeting that Huawei could begin supplying DeepSeek with training chips as early as the fourth quarter of this year.

• DeepSeek makes bold bet on training models with Huawei chips... plans to develop a model with 8 trillion parameters

Douzone Bizon set up a dedicated "AI Governance Office" to oversee companywide AI use based on the international standard ISO/IEC 42001 (AI management system). AI service platform company Wrtn Technologies selected Mirae Asset Securities as lead underwriter for its initial public offering (IPO) and NH Investment & Securities as co-underwriter. Generative AI company Xenon received approval from the Korea Exchange for a preliminary review for a KOSDAQ listing. Physical AI company RealWorld signed a memorandum of understanding with CJ Logistics to develop and commercialise a logistics-focused robotics foundation model (RFM).

• Douzone Bizon sets up AI Governance Office... led by President Yong-gu Ji • Wrtn steps up IPO push next year... selects underwriters • Xenon wins preliminary review approval for KOSDAQ listing... IPO process accelerates • RealWorld-CJ Logistics to jointly develop logistics-focused physical AI

As South Korea's generative AI services market grows rapidly, business-to-consumer (B2C) payments this year are expected to exceed 1 trillion won. With generative AI becoming an everyday digital service, including daily ChatGPT users topping 5 million, user protection and provider responsibility are also coming into focus.

• Generative AI services enter 1 trillion won era... user protection and provider responsibility emerge as tasks

As the share of corporate software budgets going to AI companies such as OpenAI and Anthropic grows, existing software companies are rolling out various countermeasures. Recently, moves to expand discounts and free benefits for AI products have stood out.

• B2B software industry's big AI discount offensive... can it break the cost burden?

Startups are accelerating efforts to develop their own models rather than using models from OpenAI or Anthropic. An analysis says the trend reflects a combination of cost-cutting, reduced dependence on major AI model developers and a growing need to strengthen control.

• Startups' in-house AI model development spreads... beyond coding, fintech also stirs