A price war among AI companies is intensifying. As Chinese AI companies expand their global market share with cost-effective open-source models, major U.S. AI companies are also responding with price moves.

Chinese AI startup DeepSeek’s new coding model V4 Flash has drawn attention for offering performance comparable to Anthropic Opus 4.8 at a very low price.

V4 Flash showed performance close to Anthropic Claude Opus 4.8 in tests for complex coding and autonomous software tasks. It is also available at a sharply lower price. Opus 4.8 costs $25 per minute of output, while DeepSeek offers its model for about 28 cents. That is about 99 percent cheaper. OpenAI cut prices for Terra and Luna in its GPT-5.6 series about 3 weeks after disclosing them. OpenAI lowered Terra fees by 20 percent and Luna fees by 80 percent, while keeping the price of its top-tier model Sol unchanged.

DeepSeek rolls out a coding AI model 99 percent cheaper than Anthropic Opus 4.8. OpenAI cuts GPT-5.6 Terra and Luna prices, targeting corporate AI cost concerns.

Interest is also growing in open-weight models that disclose model weights. Open-weight-based models also appear to be spreading in the video model segment. Chinese AI company MiniMax released a new video generation model that processes text, images, video and audio together, and has moved to compete with ByteDance and Kuaishou. MiniMax plans to 공개 the H3 model weights within days so users can download the system and modify it themselves. Many video generation models are closed, but Chinese AI companies are also seeking to expand the open-weight approach in this area.

Why is China releasing top AI for free? An 'open-weight strategy' targeting U.S. big tech. Anthropic directly rebuts controversy over banning open weights, saying control of Chinese AI is needed. "Open-weight AI is the future," Armstrong replies to Jensen Huang’s first post on X. [Tech Inside] Everything about open-weight models reshaping the AI landscape. Will video AI also see an open-weight wave? MiniMax unveils 'H3'.

With a second evaluation in August approaching for a homegrown AI foundation model project led by the Ministry of Science and ICT, participating companies are moving faster. LG AI Research released 'K-EXAONE 2.0' on the open-source community Hugging Face. The ministry and the National IT Industry Promotion Agency (NIPA) will recruit 200 people for a public evaluation panel for the 'homegrown AI foundation model' project.

LG AI Research releases 'K-EXAONE 2.0' model with 750 billion parameters. Science ministry recruits public evaluation panel for homegrown AI foundation model.

Other moves by companies in South Korea and abroad around AI are also compiled.

OpenAI CEO Sam Altman recently demonstrated a new model, 'Astra', to policymakers and regulators in Washington. OpenAI stressed that Astra can run multiple agents together for long periods to solve especially difficult tasks such as projects or advanced math problems. OpenAI’s ChatGPT is nearing 1 billion weekly active users. OpenAI launched 'ChatGPT for Academic Researchers', offering advanced AI research tools including the GPT-5.6 model family and Codex free for 1 year to researchers in science, math and engineering.

OpenAI demonstrates new model 'Astra' in Washington, saying multiple agents can operate together for long periods. OpenAI’s ChatGPT nears 1 billion weekly users. OpenAI to provide advanced AI tools free for 1 year to academic researchers.

Microsoft released its latest cases of so-called Frontier Transformation companies that create value by embedding AI into core business processes. Microsoft will also launch a super app this year that bundles Copilot chat, coding and agent functions.

Microsoft releases latest cases of Frontier Transformation companies integrating AI into core operations. Microsoft confirms Copilot 'super app' launch within this year.

Google released 'Gemini Robotics 2', a key intelligence model for implementing next-generation robots. Google also launched in South Korea a personal AI agent, 'Gemini Spark', that performs tasks in the background based on user instructions.

Google releases Gemini Robotics 2, saying it can adapt to a new robot in hours. Google launches personal AI agent Gemini Spark in South Korea.

Audio AI startup ElevenLabs expanded support channels for its enterprise AI agent platform 'ElevenAgents'. PolyAI launched a real-time voice conversation AI model, 'Dialogue-RSN-1', that directly recognizes and responds to speech.

ElevenLabs expands ElevenAgents channels, adding support such as SMS and Telegram. PolyAI launches a voice model that does not go through a switchboard, reflecting intonation and emotions.

Samsung SDS is moving forward with a multi-LLM strategy to target the enterprise AI transformation (AX) market. Samsung SDS CEO Jun-hee Lee (이준희) said on a recent earnings conference call that the company has established cooperation frameworks with OpenAI, Google Cloud and Anthropic, which it called the world’s three leading frontier AI companies. He said the ability to provide multiple LLMs optimized for customer environments is a differentiated competitiveness for Samsung SDS as the balance of performance among models is being reshaped quickly. Microsoft CEO Satya Nadella also repeatedly stressed on a post-earnings conference call that companies should not depend on a single AI lab’s model, but use multiple models together and separate the agent layer from models.

Samsung SDS CEO Jun-hee Lee says a multi-LLM strategy will strengthen AX business competitiveness. Satya Nadella says companies should move to a multi-AI-model strategy and that it is especially important for security. Satya Nadella says companies that leave everything to a single AI will find it hard to survive.

Douzone Bizon will work with Samil PwC to advance the 'ARIX model' (AI Readiness Index) that diagnoses corporate readiness for AI transformation (AX). AI-based B2B sales and marketing solution Recatch, operated by Business Canvas, will study ways to use AI for global go-to-market work with LG Chem’s CX Strategy Team and expand training on how to use it for related organizations. Wrtn Technologies will revamp the creator settlement policy of its AI entertainment content platform 'Crack'.

Douzone Bizon accelerates advancement of AI readiness index 'ARIX' with Samil PwC and others. Recatch expands AI-based global marketing collaboration with LG Chem. AI content platform 'Crack' introduces cash-out for creator rewards.

Kevin Jiang, who took part in developing Perplexity’s AI browser 'Comet', founded an AI browser startup 'Pola' targeting knowledge workers and raised $5.7 million in seed funding. Pola is focused on work automation rather than a general consumer browser.

Founder from Perplexity unveils knowledge-worker-focused AI browser 'Pola'.

Amazon Web Services, Microsoft and Google Cloud, known as the big three in the global cloud market, continued growth that exceeded expectations in the second quarter. All three made clear they would maintain an aggressive investment stance despite worries about overheating in AI infrastructure investment.

Cloud 'big three' post record growth despite AI bubble talk, keep aggressive infrastructure investment.

U.S. companies that had avoided hiring as much as possible, citing economic uncertainty and AI’s ability to cover more work, have changed. Traditional conglomerates as well as big tech companies are moving to expand hiring.

Big companies that stopped hiring entry-level workers due to AI are increasing hiring again. Why?