An engineer working for Chinese AI lab DeepSeek believes that while his firm's AI is growing at a pace that will soon replace him, he has to nevertheless continue to work on it due to competitive pressures. In a blog post, the engineer, who goes by interestingLSY/intlsy, is a kernel engineer at DeepSeek and he expresses deep skepticism for either Anthropic or OpenAI to make artificial intelligence available to the majority. Consequently, he believes that as DeepSeek's objectives align closely with making AI freely available, he has to continue working on a technology that will replace him. DeepSeek Engineer Believes […]

Read full article at https://wccftech.com/deepseek-engineer-stuck-in-self-destructive-loop-says-building-ai-to-replace-him-is-the-only-way-to-stop-openai-anthropic-from-creating-a-cyberpunk-2077-society/

Recursive Self-Improvement or RSI is widely considered the holy grail of AI, and a wild rumor now suggests that Google's DeepMind - the unlikeliest of AI labs in the current paradigm - has achieved this feat, which then puts a cynical spin on a somewhat coordinated push from Anthropic, OpenAI, and Elon Musk to try to achieve regulatory capture of the broader AI sphere by advocating for pacing progress on AI models. Cynisim abounds as Anthropic, OpenAI, and Elon Musk all call for pacing the progress on AI, which critics argue only serves as a front for disadvantaging other AI […]

Read full article at https://wccftech.com/a-wild-rumor-says-google-is-on-the-verge-of-achieving-rsi-as-anthropic-openai-and-elon-musk-all-call-for-pacing-ai-in-what-might-be-a-cynical-attempt-at-regulatory-capture/

DeepSeek's engineers are a marvel, excelling in extracting every ounce of efficiency from the architectural constraints that characterize contemporary LLMs, as they appear to have done with the just-released V4.1 Flash, which is phenomenally competitive with the likes of OpenAI's GPT-5.6 Sol and Anthropic's Claude Opus 5 on coding and cybersecurity, while entailing just a fraction of their costs! The novel architecture of DeepSeek's V4.1 Flash model DeepSeek's V4.1 Flash model sports a fairly novel architecture. Given the inherent complexity, we'll try to explain the model's main architectural elements using easy-to-understand analogies. The V4.1 Flash is a multimodal Mixture of […]

Read full article at https://wccftech.com/deepseek-v4-1-flash-beats-openais-gpt-5-6-sol-and-anthropics-opus-5-on-coding-and-cybersecurity-at-an-86x-lower-cost-while-reducing-hbm-requirements-by-3-8x-and-ssd-ones-by-8x/

When an unknown AI lab drops an anonymous model - Ox Alpha - for free, while declaring that they have the capacity to process up to 100 trillion tokens per day, we can reasonably surmise that the 'Wild West' age of cutthroat competition for AI models has arrived. Given the fact that not many AI labs have the capacity to process the declared quantum of tokens per day, the internet is going bonkers trying to figure out if the Ox Alpha is an unreleased model in the GLM family, DeepSeek's V4-Flash, or SpaceXAI's unreleased Grok model. It is quite likely […]

Read full article at https://wccftech.com/a-mysterious-ai-lab-is-offering-100-trillion-free-tokens-day-for-its-ox-alpha-model-as-evidence-points-to-zhipus-unreleased-glm/

There has been much consternation in the United States over the past few weeks around the supposed proliferation of open-weight AI models from China. Even so, DeepSeek's newly instituted peak-hour pricing mechanism belies those fears, and suggests that most of its users still reside in Asia. While this paradigm is also likely to hold true for some of China's other high-flying AI models, such as Alibaba's Qwen series of LLMs, the sheer scale of oncoming demand is truly a marvel to behold, with Qwen models alone occupying a download footprint on Hugging Face that is 2.6x that of Meta's! DeepSeek's […]

Read full article at https://wccftech.com/deepseeks-peak-hour-pricing-betrays-where-its-users-really-live-calming-us-fears-of-a-china-ai-takeover-even-as-alibabas-qwen-models-bury-meta-on-hugging-face/

DeepSeek was second only to Anthropic in terms of the total number of tokens consumed in July. And now, perhaps in a bid to cement its ascendancy, the high-flying Chinese AI lab has just unveiled the DeepSeek-V4-Pro-0813 model, its latest gambit to take on the might of OpenAI and Anthropic. The V4-Pro-0813 is now rolling out on DeepSeek API and DeepSeek Chat DeepSeek has started rolling out the V4-Pro on its API and Chat. The AI lab has priced the model at $0.435 per 1 million tokens of input and $0.87 per 1 million tokens of output. Do note that […]

Read full article at https://wccftech.com/deepseek-prices-its-new-v4-pro-0813-model-at-0-87-per-1-million-output-tokens-as-the-high-flying-chinese-ai-lab-wows-with-its-soaring-token-consumption/

Too much of a good thing can become poison, as DeepSeek appears to be finding out now after it nipped OpenAI's nascent price war in the bud by aggressively cutting the price of its V4 Flash model, unleashing a demand tsunami that its infrastructure simply can't keep up with. DeepSeek unleashed a tsunami of demand with its recent price cuts for the V4 Flash model, forcing an abrupt about-face As we detailed recently, OpenAI launched a literal price war a few days back by discounting its GPT-5.6 Luna by as much as 80 percent, with input tokens now priced at just […]

Read full article at https://wccftech.com/deepseek-forced-to-raise-prices-as-its-recent-price-cuts-to-snub-openai-unleashed-a-demand-tsnunami-that-its-20000-gpu-stash-cant-handle/

We told our readers not to discount China's uncanny ability to trounce the West in a price war, and that's exactly what its AI labs have done: merely hours after OpenAI fired its first volley, DeepSeek's revamped V4 Flash model now offers unprecedented economy and performance, while the progenitor of the Kimi K3 model, Moonshot, has just secured a sizable cluster of NVIDIA GPUs to crank up its training workloads. OpenAI now faces a dual counterpunch in its still-fledgling price war from China's DeepSeek and Moonshot As we detailed recently, OpenAI has just launched a literal price war by discounting […]

Read full article at https://wccftech.com/deepseeks-v4-flash-guts-openais-price-war-within-hours-matching-opus-grade-output-at-0-14-per-million-tokens-as-moonshot-scales-with-20000-new-nvidia-gpus/

DeepSeek's CEO, Liang Wenfeng, has disclosed a number of valuable nuggets in an investor-focused conference call, spilling the beans on China's compute deficit, the superiority of NVIDIA's GB300 GPUs, Huawei's meticulous catch-up, and DeepSeek's ongoing work on data annotation. DeepSeek's Wenfeng Says Huawei's Atlas 950 SuperPoD can fully substitute NVIDIA's GB300 NVL rack, but for every GB300 GPU DeepSeek needs 4x Ascend GPUs DeepSeek's Wenfeng disclosed a slew of valuable information in a recent investor conference call, noting that the AI lab's available compute only allows for models that activate around 10 billion parameters at any given time vs. the […]

Read full article at https://wccftech.com/deepseek-ceo-believes-nvidia-is-now-digging-its-own-grave-even-as-1-nvidia-gb300-gpu-equals-4-huawei-acend-950-gpus/

DeepSeek, the Chinese large language model developer, is said to be preparing for a 2027 IPO debut as it also looks to raise around $1.5 billion in new funds at a $71 billion valuation.

NVIDIA's H200 AI GPUs are finally shipping to China, as reported by U.S. officials, marking the return of the banned chip. China Softens Its Position on NVIDIA GPUs As First H200 Shipments Now Headed To Chinese AI Firms The NVIDIA Hopper story in China has been a roller coaster ride so far. When Hopper was launched, US export restrictions for the Chinese market were in full swing. These pushed NVIDIA to release an export-compliant variant called the H20 for China, which ended up being severely cut down, but still a viable option against what China's domestic chipmakers had to offer. […]

Read full article at https://wccftech.com/u-s-officials-confirm-first-nvidia-h200-ai-gpu-shipments-to-china/

AI chips from Huawei and NVIDIA are currently being leveraged to train and run DeepSeek’s popular models, but the days of this arrangement could be numbered as the Chinese startup has plans in motion to launch its in-house inference silicon. According to people familiar with the firm’s plans, the chip won’t be utilized for training new models, but for generating responses for users on a pre-trained one. Of course, the journey will be arduous because not only does it take years to launch a viable solution, but DeepSeek also has to contend with being limited to existing chip manufacturing equipment […]

Read full article at https://wccftech.com/deepseek-developing-custom-ai-chip-to-reduce-dependency-on-huawei-and-nvidia/

In a possible development that is almost guaranteed to raise a lot of hackles in Washington, Microsoft is considering the use of a self-hosted version of DeepSeek's V4 model for Copilot Cowork, as OpenAI and Anthropic appear determined to price themselves out of the market. Soaring token costs are driving enterprise customers away from OpenAI and Anthropic and right into the lap of China's DeepSeek For the benefit of those who might not be aware, a token is the smallest unit of data that an AI model processes, and is typically around 4-character long. A model's context window or working […]

Read full article at https://wccftech.com/microsoft-risks-trumps-ire-by-abandoning-the-costly-openai-and-anthropic-models-for-china-based-deepseeks-v4-model-for-enterprise-workloads/

Zyphra has partnered with AMD to launch its brand new open-source AI platform that rivals DeepSeek & is based in the US, all powered by MI355X GPUs. Zyphra Cloud - The DeepSeek of China Is Here & It Is Powered By An All-AMD Powerhouse With 15MW of Compute Through MI355X & Expansion To Future GPUs The Zyphra AI Cloud platform is an inference-optimized service for frontier open-weight models such as DeepSeek V3.2, Kimi K2.6, and GLM 5.1. The platform combines custom kernels, novel long-context inference algorithms, and advanced parallelism techniques to deliver high-throughput, low-latency AI performance, perfect for agentic, deep […]

Read full article at https://wccftech.com/zyphra-amd-launch-open-ai-platform-powered-by-15mw-mi355x-gpus/

The Chinese AI lab came to prominence in early 2025 after launching a large language model that trained on a fraction of the compute power and at a fraction of the cost of the big U.S. models like those from OpenAI and Anthropic.

China has unveiled its new supercomputer, called LineShine, in Shenzen, which will deliver 2 ExaFlops of compute performance. The Chinese LineShine Supercomputer Is All Set To Become The Fastest In The World, Exceeding 2 ExaFlops During a conference at the National Supercomputing Center in Shenzhen today, the fastest domestic supercomputer project was announced. Known as LineShine, the Supercomputer will be built in two phases and will deliver over 2 ExaFlops of compute. Currently, the world's fastest supercomputer is the AMD-based El Capitan, which has a peak compute of 2.8 Exaflops. Coming to the specifications, the LineShine Supercomputer is purely a […]

Read full article at https://wccftech.com/chinas-lineshine-supercomputer-2-exaflops-47000-cpus-zero-reliance-on-foreign-chips/

DeepSeek says both models are more efficient and performant than DeepSeek V3.2 due to architectural improvements, and have almost "closed the gap" with current leading models, both open and closed, on reasoning benchmarks.

Chinese artificial intelligence lab DeepSeek claims to significantly reduce computing resources required for token inference and memory resources with its latest V4 model, according to its release notes. DeepSeek claims that the V4 AI model requires just 27% single-token inference FLOPs and 10% of key-value (KV) cache when compared to its predecessor, the DeepSeek V3.2 model. The reduction in cache requirements addresses memory requirements, with lower requirements conserving memory and increasing the context available to model builders when creating their models. How DeepSeek V4 Slashes Compute and Memory Costs In its release notes for DeepSeek V4, DeepSeek outlines that the […]

Read full article at https://wccftech.com/deepseek-v4-cuts-kv-cache-by-90-at-1m-tokens-but-aggressive-compression-could-risk-needle-in-a-haystack-failures/

Anthropic accuses DeepSeek, Moonshot, and MiniMax of using 24,000 fake accounts to distill Claude’s AI capabilities, as U.S. officials debate export controls aimed at slowing China’s AI progress.