Alibaba

Someone posted the proof yesterday and it's been sitting with me since. Higgsfield's new Genjutsu Lite model, going by screenshots of their own API docs, is Alibaba's Wan 3.0 wearing a different name. Genjutsu is Seedance 2.5. Same…

China's AI stocks are getting HIT in Hong Kong today. The Hang Seng Tech Index fell about 4% to around 3,834, while the Hang Seng Index dropped almost 3%. According to Chinese media, AI model makers MiniMax and Zhipu fell more than 4%,…

Wild. Why we can't get cheap NVIDIA cards. https://t.co/frFkm8HW3A

This is absolutely INSANE and another example of just how EXTREME demand for AI compute has become. According to Reuters, China's three largest technology companies, ByteDance, $BABA, and Tencent, have reportedly placed more than $35B in…

ByteDance, Alibaba, Tencent place over $35 billion worth of Nvidia chip orders US-based chipmaker Nvidia has told customers and partners it cannot guarantee that it can fill the ‌orders because demand from Chinese technology companies far…

Alibaba’s RISC-V IP powers over 6 billion shipped chips as tech giant targets AI The milestone comes as Chinese chipmakers seek to reduce dependence on foreign tech amid US-led export restrictions https://t.co/Mje2xNdkyJ via @scmpnews

Alibaba just said more than 6 BILLION chips have shipped using its XuanTie RISC-V designs. Damo Academy chief scientist Meng Jianyi announced the number at the RISC-V Summit China in Shenzhen. That is up from about 4 billion back in 2024.…

Alibaba Damos/T-Head team unveiled E908A auto grade RISC-V core. Reached both ISIL-B and ISIL-D standard RISC-V CPU IP. It is also getting next-gen C950A ready for auto chip mkt. T-Head team has delivered 6B+ RISC-V chips. Auto grade is an…

IDC's estimate on China's AI accelerator mkt in Chinese AI Data Center in 2026H1: Take this for what you will, but it estimates 2m total chip & 33.8% for Nvidia & 28.5% for HW. Some # here look very questionable to me like Alibaba T-Head.…

I think Gavin's correct. Open-weight models that compress frontier margins are positive for AI infra -- they don't shrink demand for compute. An example is after Deepseek R1, Tencent, Alibaba and Bytedance *significantly* increased H20…

No tokens. Just a number between 0 and 1. That's the entire output of Liquid AI's d1 family, and this week they shipped two new ones, open weight this time: d1-3B for text and vision, d1-omni-600M for text plus image or audio, both aimed…

AI: Nvidia’s Open Source ‘AI Factory’ Push vs China. AI-RTZ #1233 (Part 1)…Nvidia the Kingmaker, part 1: Reflection’s Beam arrives, Jensen’s latest answer to China’s open source lead.$NVDA https://t.co/Daf67mIt6o https://t.co/Kws4gcYFh7

China is about to unleash a 60x AI demand shock and it could create an entirely new group of stock market winners. JP Morgan expects China’s inference-token consumption to grow roughly 60-fold between 2026 and 2030. The chart shows demand…

A closer look at what we brought to #ROSCon Global 2026 with reBot. A few demos from the booth: ✅#Qwen3-VL Skills — vision-language reasoning with reBot ARM ✅ #NVIDIAIsaac — reBot teleoperation in simulation ✅ Closed-loop Policy —…

If you're a robotics researcher working with GPT 6 Astra to do interesting things with robotic control or agentic real2sim, and would like access to open source models (Kimi K3, Qwen 3.8 Flash Next) as a comparison, please DM me! I'd love…

Run #LLMs on @NVIDIARobotics #Jetson without x86 ONNX export. 🚀 #TensorRT-Model-Connect on reComputer Classic J5012 (#AGXOrin 64GB, #JP7.2). @huggingface checkpoint → trtmc build → .bundle on device trtmc run for text gen Tested:…

The pattern I keep seeing this month: once a model layer commoditizes, the fight moves one layer down to whoever serves it fastest. RunAnywhere, a YC W26 company, launched Wally this week, an inference stack for open frontier models that…

My read on Xiaomi shipping MiMo-V2.6-Pro this week: the real story isn't the benchmark score. It's who is now capable of putting up that score. Xiaomi, a phone and appliance company, published MiMo-V2.6-Pro and Flash on Hugging Face…

Another Chinese lab just gave away a trillion parameter model for free, and the market barely blinked because it's the fourth time this year. Xiaomi open sourced MiMo-V2.6 today: two natively omnimodal models, Pro and Flash, both MIT…

a few notable open model releases *since* this interview of @dylan522p by the way :) Thinking Machines - Inkling - Jul 15 Moonshot - Kimi K3 - Jul 16, weights Jul 26/27 inclusionAI - Ling 3.0 Flash - Jul 23 DeepSeek - V4-Flash-0731 - Jul…

The shift I keep coming back to this week: not every AI call needs to write a sentence. Diogo Almeida's Jev proved that in nine days. Jev is what TypeSafe calls a System One model. It skips token by token generation entirely. Feed it a…

Apple shipped something small this cycle that I think matters more than another point release of a frontier model: macOS 27 now includes an fm command in the terminal, putting Apple's on-device model one shell command away from every…

The gain came from the harness, not the model. That line from https://t.co/VlWgutdPl7's research roundup today sums up where agentic engineering is right now: 41.2% to 91.8% success on long-horizon robot tasks, same Qwen3-8B checkpoint,…

Frontier agentic performance just hit $1 per million. StepFun launched Step 5 Preview today. 600B total parameters, 27B active per token. 1M context. $1 per million input tokens. That is 10 to 20x cheaper than GPT-6 Astra at frontier…

Open weights image generation just became serious. Qwen-Image-2.1 ships today. Generation and editing in one model. 2K resolution native. Professional typography. Alpha channel. Open weights. The market map: GPT Image delivers impressive…

1/5 LLMs fail long-horizon tasks for a fixable reason. 2/5 Plans look valid until they run. A robot reaches for an object that was already moved. A step gets skipped because its precondition was never checked. The LLM had no model of how…

1,424 tok/s prefill on consumer hardware. Halogen is a dedicated inference engine for Qwen3.8-Flash-Next on Strix Halo. 4x faster than previous implementations at 32K context. Local AI performance is now a software optimization race. The…

27B model. 5.9GB. 98% of full performance. PrismML compressed Qwen3.8 27B by 9x using ternary weights with almost no accuracy loss. Biggest gains in agentic coding and long-horizon tool use. 262K context. Fits on 8GB VRAM. Apache 2.0.…

Qwen3.8-27B hits 381 tok/s locally. Here's how: → DFlash2 kernel: 138 tok/s baseline → Optimized MTP: parallel multi-token prediction → Context lookup: model verifies 16 tokens at once → Prefix cache: TTFT drops from 22s to 0.56s Ceiling…

381 tokens per second on one GPU. That's Qwen3.8-27B running via DFlash2, context lookup, optimized MTP, and prefix cache. The 381 only appears in a specific regime: when the answer already lives in the prompt, the model verifies 16 tokens…

Open model adoption just hit a tipping point. Here's what that means for builders: 1/5 - The Volume Evidence 78.4% of tokens on Vercel's AI gateway run on open models. That's not a survey. That's production traffic. The switch is already…

Open source just shipped a 744B agentic model. Shanghai AI Lab released Atria Dawn Preview on September 11 with no blog post, no pricing announcement, and no press release. Just a Hugging Face commit and an FP8 checkpoint the next day. The…

Frontier AI now fits in your pocket. PrismML shipped Ternary Bonsai 2 27B this week. 27.8 billion parameters. 5.9 gigabytes. That is 9x smaller than the full-precision Qwen3.8 27B it is based on. The retention number is what surprised me.…

Qwen3.8-Omni-Flash is a different bet. Not "better language model". Omni-modal from the ground up. One model that handles: → Text, image, audio, video in → Text and audio out → Real-time inference at flash speeds The architecture isn't a…

A new agentic open-weights model drops Tuesday. 3x cheaper than DeepSeek Flash. 15% better on long agentic loops. Available on RouteLLM API and on the coding agent. Designed for 90% of coding tasks. My read on what's happening in the…

🔥🔥#RK1828 vs. #RK1820: 32K context. Fully local. We pushed long-context AI inference further on the #reComputer RK3576 Dev Kit with #Qwen3 running locally: ⚡ RK1828 + Qwen3-4B → 32K NIAH: 12/12 PASS → 74.75 tok/s generation → ~300 ms…

AI & Tech Brief: Exclusive | Ro Khanna seeks to pace Chinese frontier AI The top Democrat on the CCP select committee is asking DeepSeek and Alibaba to pace frontier AI https://t.co/Z1PMXgxJLB

Agentic AI is moving to the edge. On NVIDIA Jetson AGX Thor, TensorRT Edge-LLM completed MLPerf Edge Agentic 6.4x faster than the llama.cpp reference, reaching 52.33 token/sec for Qwen 3.6 27B model. NVFP4, ~96% of prompt tokens served…

Fathom lets each query decide how many bits of each key channel it reads. At 1M tokens on Qwen3-8B a decode step is 1.67x faster than 136-bit scans, and it matches the most accurate 136-bit scan at 92 bits on coding-agent sessions. No gain…

Four Mac Studios fine-tuning Qwen3-9B over RDMA-over-Thunderbolt reach 936 tokens/s at sequence length 17408, up to 1.6x the single-trunk, non-overlapped baseline. Measured Thunderbolt bandwidth is far below the nominal spec. Unified…

🔥🔥RUN #7B #VLM on reComputer #RK3576 Dev Kit + #RK1828 We tested #Qwen 2.5-VL-7B_prune on the reComputer RK3576 Dev Kit + RK1828 AI Accelerator: ⚡ TTFT: 418.93 ms 🚀 Streaming Output: 330.68 chars/s ⏱️ End-to-End: 2.84 s From image…

Double the throughput on the same edge platform. ⚡ See Qwen3.5 9B NVFP4 run with DFlash speculative decoding turned on and off, reaching 68.1 tokens/s with DFlash enabled on NVIDIA Jetson. https://t.co/VeDRiQDmz8

Jon Cipher is right about the structural shift here. The OpenRouter data tells a clear story. Chinese models took the lead in OpenRouter token share in March and kept it. Not because they win every benchmark. Because restricted compute…

🔥🔥Testing 7B #LLM on #RK3576+#RK1828 We put Qwen2.5-7B (W4A16) to the test on the reComputer RK3576 Dev Kit with RK1828 AI Accelerator: ⚡ Prefill: 121.12 tok/s 🚀 Generation: 60.42 tok/s ⏱️ TTFT: 309.64 ms But this isn’t just another…

The inference layer just got a native desktop. Cline shipped Cline Desktop this week: a standalone app for open-weight models, not just a VS Code extension. ClinePass at $9.99/month gets you DeepSeek V4, GLM 5.3, Kimi K3, Qwen3.7, and a…

Coding agents just split into two permanent tiers. Cline Desktop shipped today: native macOS interface for working with open weights models. ClinePass provides access to DeepSeek-V4.1-Flash, Musespark-1.3, and others at no API cost. BYOK…

Anthropic's Sept threat report: a China-based actor built about 16 air-defense-suppression modules with Claude, one scenario listing 12 Taiwan targets; per the report Alibaba ran 151 million distillation queries. YC's Garry Tan says US…

Local inference is becoming a product category, not just a benchmark table. Perplexity launched Portable Computer on Windows RTX PCs today. The orchestrator LLM, subagent LLM, and the full agent harness all run on-device. No cloud…

Run Qwen3.8 Flash Next, a 125B multimodal MoE model, locally on NVIDIA Jetson AGX Thor. ⚡ Built for reasoning, coding, and tool use, its 262K context window can handle large codebases and long conversations. With NVFP4 and MTP, it reaches…

Most code review tools today are either fully LLM-driven or fully rule-based. Alibaba's open-code-review is neither. They open-sourced the tool they run at scale internally. The architecture is hybrid: a static analysis pipeline handles…

👁️ Testing Qwen2.5-VL-3B on the reComputer RK3576 Dev Kit with RK1820 AI Accelerator ⚡ 361.98 ms Time to First Token 🚀 227.56 chars/s Streaming Output 🧠 Local Vision-Language Model 🔍 Image Understanding · Visual Q&A 🔒 Fully Local…

HSBC: Chinese hyperscalers capex spend estimates as % of revenue is far lower than US hyperscalers. $BABA $BIDU $AMZN $META $GOOGL $META $ORCL

Anthropic says Chinese labs routed at least 35 million user requests to Claude over the summer via thousands of fake accounts, naming DeepSeek, Moonshot and Alibaba. This is Anthropic's account; the labs have not confirmed. Distillation is…

On an 8x B200 node serving Qwen3-Coder-480B, Nvidia's Max-Q profile bought 8.6% more tokens per joule at a 5.2% latency cost. A calibrated per-lane setting gave 20.4% at 3.5%. Three days of running saved 32.3% of a lane pair's electricity.…

From Anthropic's new distillation accusations against Chinese AI labs: - Fable wasn't used for distillation. Zai tried but failed. This means Anthropic can stop distillation with sufficiently strong guardrails? Opus was used. - It's not…

HyQuant keeps the top ~5% of key positions and a 128-token window in FP16 and quantises the rest to 4-bit: decode kernel up to 3.58x FlashAttention-2 at 32K context, LongBench 45.04 vs 44.59 on Qwen3-8B. Selective precision beats uniform…

🔥🔥RK3576+RK1820: 178.86 tokens/s. Local LLM, Running Fast. We put the #reComputer #RK3576 Dev Kit with #RK1820 AI Accelerator to the test. Running #Qwen 2.5-0.5B, it achieved an impressive 178.86 tokens/s in our test. 🧠 Rockchip RK3576…

A 2.4T-parameter Qwen3.8 with 95B active per token now fits one node: 8 B300 GPUs, NVFP4 quantization to about a third of full size, over 4,000 tokens/s per GPU, a 1M-token context and a low/medium/high reasoning budget. Open weights only…

Zai & Moonshot need to speed up their releases. DeepSeek took a while to get this top notch V4 architecture up & their CPT & post training pipeline since then has been great. Ppl like me will move exclusively to DS if it's this much…

Alibaba’s AI ‘employees’ can now work in ByteDance, Tencent apps https://t.co/I1jL52AQ4V

Qwen3.8 27B at 4-bit (17GB) matches the 55GB BF16 model on Terminal-Bench 2.1, fitting a 24GB RTX 4090. 2-bit (10.7GB) slips. 1-bit (6.2GB) scores at random chance on GPQA Diamond, and longer reasoning makes it worse. Quantization damage…

NSA, FBI and CISA jointly accuse six Chinese AI firms — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, https://t.co/CaUaiwWiiM — of industrial-scale distillation of US frontier models via APIs, clouds, aggregators and reseller 'transfer…

I think the point that "astra is the first vlm that doesn't suck" is a really important one. At the smaller scale the qwen models are incredible but you feel the limitations. Still waiting on my qwen 3.8 8b/14b.... https://t.co/Qa2xif9sWG

"China's four major cloud service providers (CSPs), Baidu, Alibaba, Tencent, and ByteDance, are strongly requesting Long-Term Agreements (LTAs) from Samsung Electronics and SK Hynix" 👀👀👀 $DRAM $DISK $MU $EWY https://t.co/cQ8WTpKlQn

Open weights are not the same as cheap serving. Alibaba's Qwen 3.8-Max, 2.4T parameters with 95B active, is estimated to need 48-64 B200-class GPUs to serve. DeepSeek's V4 Flash, 284B, fits in about 142GB at FP4. Downloadable and…

And then DeepSeek, Moonshot, Zhipu, Alibaba...then Tencent, Bytedance, Minimax, Xiaomi.... https://t.co/QlG3ehFkDq

‼️ WHAT JUST HAPPENED?! China’s Richest People Have One Problem Money Can’t Fix China has plenty of billionaires, but wealth doesn't automatically buy political power. The Communist Party remains firmly above the country's richest people,…

The AI assistant category is being replaced. QwenWork landed in public beta globally this week. Alibaba's move is to make the next product category a single subscription that puts an AI agent in control of your browser, your desktop, and…

Post Log in Sign up Post Chubby♨️ on X: "wtf Qwen 3.8 has been updated to version 0902 and is now almost at the same level as Fable 5 in benchmarks. The interesting two aspects: 1) For such significant jumps, it seems that you don't even…

Vytautas Valinskas / Technology.org (Aug 31, 2026), largely recapping TechCrunch Aug 28: Xpeng robotics unit raised more than $900 million at post-money valuation above $6.3 billion (company: largest single private financing in China's…

Some AI/semis notes from the past week: -> $NVDA huge Q2 revenue and profit growth. 2027+ is more growth even under a bottlenecked environment. Nvidia GOAT company? -> $NBIS first cloud customer for $NVDA Groq 3 LPX inference rack which…

Startup Fortune (Aug 29, 2026, 9:38 PM): Citing Fortune, Unitree raised about $900 million on Shanghai STAR at a roughly $9 billion IPO valuation, then closed Aug 19 first day up more than 460%. Backers named: DeepSeek, Alibaba, Ant Group,…

XPENG is putting serious money behind humanoid robotics. Its robotics business just raised more than $900 million, reaching a post-money valuation above $6.3 billion. The round was led by IDG Capital, with Tencent, Alibaba and Gaorong…

Story 2, Shanghai. Unitree is down 45%+ from its IPO peak. Not a robot story. A supply story. Alibaba's record share sale and YMTC's IPO plans landed the same week. The cost of the AI build-out just got handed to new shareholders.

XPeng's robotics unit raised over $900M at a $6.3B post-money valuation. Largest single-round private raise in China's embodied AI sector. IDG Capital led. Tencent, Alibaba, and Gaorong joined. Worth understanding the structure. XPeng Inc.…

The Robot Report Continue to Site Facebook X LinkedIn Reddit Pinterest Share XPeng’s Dogotix unit is developing the IRON humanoid robot. Source: XPeng Like other global automakers, XPeng Motors is branching into humanoid robots. The…