iFlytek
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 8
- iFlytek to unveil Ascend 950-trained flagship LLM at 1024 event
iFlytek plans to announce a new flagship general-purpose large model trained on Huawei's Ascend 950 at its October 24 1024 Developer Festival, a bid to escape the domestic-compute bottleneck. The article says Spark has been the only mainstream LLM trained entirely on Chinese chips for three years, but Ascend 910B's memory and bandwidth left its long-context and complex-code training about 5x less efficient than Nvidia's H200; Ascend 950's larger memory and bandwidth raise theoretical efficiency against H200 to 65%–85%, giving Spark a base for top-tier models. Commercially, the piece notes H1 2026 revenue of RMB 11.623 billion, up 6.52% year on year, a RMB 204 million net loss attributable to shareholders, a 2.65% decline in government-business revenue, and chairman Liu Qingfeng's August admission that iFlytek was visibly absent from the code and agent boom. The event is also expected to cover funding progress for the spun-out Ciyuan Xinghuo unit and chip and supernode plans.
36氪 人工智能 · 🔥 6 - Alibaba, Tencent and ByteDance race into AI office hardware to capture context
Alibaba, ByteDance and Tencent are competing over AI office hardware, treating recording cards, AI earbuds and smart glasses as entry points for collecting offline "context", according to a China Entrepreneur Magazine report republished by 36Kr. Tencent's WorkBuddy opened its platform in September with more than 100 ecosystem partners and integrations into dozens of devices from Plaud, Rokid and Insta360, while Alibaba's Qwen Office launched the QwenNote A2 AI agent device on Sept 22 for 1,199 yuan and followed on Oct 13 with N1 and N1 Pro AI glasses plus clip-on earbuds. ByteDance's hardware matrix includes Doubao earbuds, a Doubao phone and the AI recording bean developed with Anker Innovations under Feishu, as investors and practitioners cited in the report warn that companion AI hardware has very high idle rates while funding has cooled and customer acquisition costs remain steep.
36氪 人工智能 · 🔥 3 - Analysis: AI's academic roots, concentration and going global
A column by Yan Deli of Tencent Research Institute, republished by Cyzone, traces AI's history: the 1943 M-P neuron paper, the 1950 Turing test and the 1955 Dartmouth proposal that named the field, which together answered how to express intelligence in computation, how to evaluate it and how to create a new discipline. The piece argues academia and industry are deeply intertwined—AlexNet (2012) and Transformer (2017) as the key papers behind deep learning and today's large language models, and OpenAI's 2020 scaling-law paper as the guide for model development—with AI firms following a chip-talent-model-paper loop. It cites third-party figures for high concentration: Stanford's AI Index Report 2026 counting 59 notable AI models from the US and 35 from China; The Information finding Anthropic and OpenAI together took 89% of annualized revenue across 34 leading generative-AI startups; and 1,112 generative AI services filed in China as of Aug. 31, with Beijing at 26.8%. It also says Chinese AI firms internationalized early, citing Manus, MiniMax and Kuaishou's Kling as overseas-revenue focused.
创业邦 科技 · 🔥 0 - iFlytek to unveil Spark X2.6 and cloud-edge plan at Oct 22 developer festival
iFlytek said its 2026 global 1024 Developer Festival will open on October 22, where it plans to release iFlytek Spark X2.6 along with a new "cloud + edge" collaboration solution. The company said its "Token + Agent" Spark developer ecosystem will also be upgraded. The event will also showcase AI applications in education, healthcare, office and automotive scenarios.
36氪快讯 · 🔥 0 - Tencent launches Hy Translate app in 12 markets on Hunyuan Hy-MT2
Tencent released the Hy Translate app the day before Mid-Autumn Festival, built on its Hunyuan Hy-MT2 model and launched in 12 markets including the US, Japan and South Korea; it supports 33 languages plus five Chinese minority languages and dialects (Cantonese, Tibetan, Mongolian, Kazakh and Uyghur). A hands-on review says users can switch among academic, business, colloquial, fiction, marketing and news styles and add custom instructions, and that a single roughly 1GB offline model package covers all 33 languages, though simultaneous interpretation and other advanced features are missing and its split-screen conversation UI is less polished than iFlytek's. The article notes Youdao already offers polishing presets via DeepSeek-V4 but rates its flexibility below Hy Translate's conversational customization.
钛媒体 · 🔥 0 - Global AI Chip Summit 2026: from TOPS to tokens as agents reshape silicon
The 2026 Global AI Chip Summit in Shanghai brought together more than 60 speakers across an opening session and three forums on large-model AI chips, agent inference chips and embodied-intelligence chips, focused on the computing chain from TOPS to tokens — covering architecture innovation, 3D stacking and near-memory compute, KV cache, new memory types, compute-in-memory, analog computing, RISC-V and token-factory operations. Several vendors used the stage to preview or detail products: Houmo AI said its M50 chip has reached 150+ design wins in half a year of mass production and that its next flagship A20 has taped out; Intel previewed a high-density cabinet with 128 or 192 Granite Rapids processors that it says can run 50,000 concurrent agents; Maitech said its 3D TPU has taped out and is expected back in November. Those figures are vendor claims and have not been independently verified.
品玩 实时要闻 · 🔥 0 - Nikkei: China's major AI firms launched 16 new models in September
Nikkei data reported by TechNews shows 10 major Chinese AI companies — including DeepSeek, Alibaba, Tencent and Xiaomi — launched 16 new models in September, mostly lightweight and low-cost, such as DeepSeek V4.1-Flash, Zhipu's GLM-5.3-FlashX and MiniMax M3.1-Flash-Preview. BenchLM counted 20 large language models from eight major Chinese firms in July–September, up from 2 in the first quarter and 17 in the second, while the anonymous model Space Bunny Alpha — believed to be a stealth MiniMax release — surged in usage on OpenRouter. The report contrasts this with Anthropic CEO Dario Amodei's calls to slow AI R&D and crack down on illegal distillation, noting Chinese firms increasingly train and run models on domestic chips.
科技新报 · 🔥 0 - AI's sci-fi PR problem: from Claude's enzyme claim to a '$100 trillion ecosystem'
A commentary republished by Huxiu argues that many of the AI industry's biggest "breakthroughs" happen in press releases rather than in the lab. It cites Anthropic's claim that 950 Claude agents worked for 21 hours and burned 210 million tokens to sift more than 200,000 reverse transcriptases for a previously undescribed enzyme system named ART — yet the paper is still a preprint, the system's function is unknown, and rerunning the same task found the key sequence only once in 10 attempts. The piece also lists Zhipu's talk of a path to recursive self-improvement, Dreame's pledge to build a "$100 trillion ecosystem," and UBTech's fastest-growing financial category being smart litter boxes and pool cleaners.
虎嗅 AI · 🔥 0
Experience and discussion from the community
Share my iFlytek experienceAsk about iFlytek
Nobody has shared their experience with iFlytek yet.