Tags: DeepSeek

DeepSeek founder Liang Wenfeng told prospective investors that the Hangzhou‑based lab will prioritize artificial general intelligence and open‑source models over short‑term revenue. The Chinese AI startup is seeking at least $300 million in its first outside capital raise, targeting a 70 billion‑yuan (~$10 billion) valuation. Previously funded solely by Liang’s hedge‑fund High‑Flyer Quant, DeepSeek’s shift reflects the scale of its training runs and a strategic move toward the domestic market. Investors remain unnamed, and the company declined comment as regulators watch the sector closely. Ler mais

Beijing‑based Moonshot AI closed a $2 billion financing round led by Meituan’s Dragon Ball arm, with China Mobile, CITIC Private Equity and other investors participating. The deal values the Kimi chatbot maker at over $20 billion—roughly seven times its December 2024 valuation. Founded in 2023 by three Tsinghua alumni, Moonshot has seen its consumer‑facing Kimi product double annual recurring revenue to more than $200 million in two months. The funding surge places the firm among China’s most heavily backed AI labs and fuels speculation about a near‑term public listing. Ler mais

Chinese AI lab DeepSeek is in talks to raise its first venture‑capital round, aiming for a valuation that could climb from $20 billion to $45 billion. Founder Liang Wenfeng, who controls about 90% of the company, is turning to investors after rival firms began poaching key researchers. The round is expected to be led by the state‑backed China Integrated Circuit Industry Investment Fund, with cloud giants Tencent and Alibaba reportedly considering participation. DeepSeek’s efficient large‑language model runs on Huawei chips, positioning it as a strategic asset in Beijing’s push for AI independence. Ler mais

A new Appfigures report shows that releasing image‑generation models has become the most effective way for AI mobile apps to attract users, delivering up to 6.5 times more downloads than traditional updates. OpenAI's GPT‑4o image model, Google’s Gemini Nano Banana, and Meta AI’s Vibes each sparked multi‑million install spikes, though only OpenAI turned the surge into significant consumer spending. The findings suggest visual capabilities now trump chat‑only upgrades in driving app growth, but the revenue impact remains uneven across providers. Ler mais

In a California federal courtroom on Thursday, Elon Musk told a judge that his AI startup xAI employed OpenAI’s models to develop its own system, Grok, through a practice known as model distillation. Musk said the technique is common across the industry, answering “partially” when asked if xAI directly distilled OpenAI technology. The testimony highlights a growing debate over the legality and ethics of AI model sharing, with companies like OpenAI, Anthropic and Google warning of potential intellectual‑property violations. Ler mais

DeepSeek announced a 75% promotional discount on its new V4‑Pro model and reduced cache‑hit charges across its entire API to 10% of previous rates. The price cut, effective immediately and running through May 5, 2026, makes the model cheaper than OpenAI, Anthropic and Google offerings even at full price. The move intensifies a pricing battle amid U.S. accusations that Chinese firms are distilling American AI models at scale, positioning DeepSeek as a low‑cost alternative for developers and enterprises. Ler mais

Chinese AI firm DeepSeek released two new large language models, V4‑Pro and V4‑Flash, both featuring a one‑million token context window and open‑source licenses on Hugging Face. V4‑Pro, a 1.6‑trillion‑parameter model, outperformed leading U.S. models in coding and agentic tasks, while V4‑Flash delivered comparable speed at a fraction of the compute cost. DeepSeek also announced a token price of $3.48 per million output tokens, dramatically undercutting OpenAI and Anthropic rates, positioning the models as cost‑effective alternatives for developers. Ler mais

OpenAI rolled out GPT‑5.5 for paying users, DeepSeek previewed its V4 series and Anthropic launched Opus 4.7, intensifying a three‑way race to dominate the next generation of artificial‑intelligence tools. The releases target coding, reasoning and agentic tasks, each promising longer context windows, cheaper hardware deployment and sharper output quality. While OpenAI highlighted the model’s intuitive coding assistance, DeepSeek touted a hybrid‑attention architecture that retains long query histories. Anthropic’s Opus 4.7 focuses on literal prompt interpretation and refined visual aesthetics. The flurry of announcements underscores a fast‑moving market where companies vie for developer and enterprise adoption amid geopolitical scrutiny. Ler mais

Chinese AI lab DeepSeek released two preview versions of its next‑generation large language model, DeepSeek V4 Flash and V4 Pro. Both models use a mixture‑of‑experts architecture and support a 1‑million‑token context window, enabling users to feed entire codebases or long documents into prompts. DeepSeek says V4 Pro, with 1.6 trillion parameters (49 billion active), is the largest open‑weight model on the market, while V4 Flash offers a smaller, more affordable option. The company claims the new models narrow the performance gap with leading closed‑source systems and are priced well below competing frontier models. Ler mais

DeepSeek unveiled two new AI models, V4 Pro and V4 Flash, promising a context window of up to one million tokens and open‑source access. The company claims the Pro version rivals top closed‑source systems in reasoning, while the Flash variant offers faster responses with comparable performance on simple tasks. Shortly after the release, U.S. federal agencies barred the app from government devices, citing national‑security concerns, and South Korea paused downloads over privacy issues. The moves highlight a clash between rapid AI innovation and emerging regulatory scrutiny. Ler mais

Nvidia chief Jensen Huang told listeners on the Dwarkesh Podcast that DeepSeek’s plan to run its upcoming V4 foundation model on Huawei’s Ascend 950PR processor would be “a horrible outcome” for the United States. The Chinese lab’s migration from Nvidia’s CUDA software to Huawei’s CANN framework threatens the hardware‑software dependency that has underpinned America’s AI dominance. Huang’s remarks come as U.S. lawmakers consider adding DeepSeek to the export‑control entity list, and as the industry watches whether Huawei’s chips can close the performance gap with Nvidia’s GPUs. Ler mais

Anthropic announced that, effective 3 p.m. ET on April 4, its Claude AI will no longer be free for third‑party applications. Users of OpenClaw and similar tools must now purchase a usage bundle or provide a Claude API key. Founder and head of Claude Code, Boris Cherny, cited engineering constraints and capacity limits as the reason for the change, noting that existing subscription plans were not designed for the heavy usage patterns of these integrations. The move forces developers and end‑users to reconsider how they access Anthropic’s models. Ler mais

Anthropic has warned that three Chinese artificial‑intelligence firms—DeepSeek, Moonshot and MiniMax—conducted large‑scale campaigns to illicitly extract capabilities from its Claude chatbot. The company says the firms used roughly 24,000 fraudulent accounts to generate more than 16 million exchanges, effectively using Claude as a shortcut to improve their own models. Anthropic cited IP address data, metadata requests and infrastructure clues to link the activity to the companies with high confidence. The firm plans to upgrade its systems to make such attacks harder and easier to detect, while noting similar concerns raised previously by OpenAI. Ler mais

DeepSeek, in partnership with Peking University, unveiled Engram, a new training method that separates static memory from computation in large language models. By using hashed N‑gram lookups and a context‑aware gating mechanism, Engram reduces reliance on high‑bandwidth memory (HBM), allowing models to operate efficiently on standard GPU memory while scaling parameter counts. Tests on a 27‑billion‑parameter model showed measurable gains across industry benchmarks, and the approach integrates with existing hardware solutions such as Phison’s SSD‑based accelerators and emerging CXL standards. Engram could ease pressure on costly memory hardware and stabilize DRAM price volatility. Ler mais

OpenAI's dominance in the generative‑AI market has waned as competitors such as Google, Anthropic and DeepSeek have introduced new models that outpace its latest release, GPT‑5. The company, led by Sam Altman, has responded with accelerated development and internal restructuring, but analysts note that its reliance on external funding and costly infrastructure deals leaves it vulnerable. While ChatGPT still draws hundreds of millions of users, the rapid growth of rival platforms and the escalating cost of AI‑related hardware have intensified a "code red" atmosphere within OpenAI. Ler mais

Nvidia chief executive Jensen Huang cautioned that China can construct AI data centers and even hospitals far faster than the United States, citing the country's expansive energy resources and swift construction capabilities. At the same time, a report from OpenRouter and Andreessen Horowitz shows Chinese open‑source large language models now account for roughly 30% of global AI token usage, up from just over 1% a year earlier. While Huang affirmed Nvidia’s chip technology remains ahead of China, the rapid growth of Chinese AI models and the nation’s infrastructure advantages highlight an intensifying competitive landscape. Ler mais

In a week‑long AI‑only poker showdown, OpenAI’s o3 model emerged victorious, out‑earning the other eight large‑language‑model competitors. The contest featured nine chatbots—including Anthropic’s Claude Sonnet 4.5, X.ai’s Grok, Google’s Gemini 2.5 Pro, Meta’s Llama 4, DeepSeek R1, Moonshot’s Kimi K2, Mistral’s Magistral, and Z.AI’s GLM 4.6—playing thousands of hands of no‑limit Texas hold ’em at $10 and $20 tables with $100,000 bankrolls each. While the bots displayed strong strategic play, they struggled with bluffing, position, and basic math, highlighting both progress and lingering gaps in AI decision‑making under uncertainty. Ler mais

Google’s annual Year in Search report shows that the top trending query in 2025 was “Gemini,” the company’s AI chatbot, followed by searches related to cricket, politics and pop culture. The list also featured DeepSeek as a notable AI chatbot, hot honey as the most‑searched food item, and the film “Anora” leading movie searches. Sports enthusiasts gravitated toward the FIFA Club World Cup and teams such as Paris Saint-Germain, while podcasts and books saw “The Charlie Kirk Show” and Colleen Hoover’s “Regretting You” at the top of their categories. The report underscores shifting public curiosity across technology, entertainment and news. Ler mais

Amazon Web Services introduced a suite of new capabilities aimed at simplifying the creation of custom large language models for enterprise customers. At its re:Invent conference, AWS unveiled serverless model customization in SageMaker, offering both point‑and‑click and natural‑language‑driven workflows, and announced reinforcement fine‑tuning in Bedrock. The company also launched Nova Forge, a service that builds bespoke Nova models for a fixed annual fee. These moves signal AWS’s focus on frontier AI models and could help customers differentiate their AI solutions in a market dominated by Anthropic, OpenAI, and Gemini. Ler mais

Chinese startup DeepSeek has released two new AI models—DeepSeek‑V3.2 and DeepSeek‑V3.2‑Speciale—under an open-source license. The models claim performance comparable to GPT‑5 and Gemini 3 Pro on long‑form reasoning, tool use, and dense problem solving while offering a 128,000‑token context window and reduced computational cost through Sparse Attention. Their launch challenges the dominance of U.S. AI firms, sparks regulatory scrutiny in Europe, and raises questions about the future of AI accessibility and geopolitics. Ler mais