DeepSeek

95.4% accuracy and $5.63 per 1,000 correct tool calls sits right next to 95.4% accuracy and $1.50 per 1,000 correct tool calls on the same benchmark. Same task, same pass rate, 3.75x the bill. That's Claude Opus 5.5 and Grok 4.7 on Tirtha…

‘Jensen’s nightmare’: Huawei, DeepSeek narrow AI chip ecosystem gap with Nvidia DeepSeek is likely to rely ‘primarily’ on Huawei’s ‘already mature and well developed’ technology stack as domestic semiconductor industry gains steam at…

I swear, we do this every 3 months. "Chinese models are falling hopelessly behind. They don't have enough compute". And then they release their new models "Oh, they are caught up. Need to ban them". Why not just recognize when flagship…

Implication of such heavy hitters investing in DeepSeek is not only DS being used by these corporations on a large scale in their physical or digital products, but also additional data coming in for DS + more importantly, DS has access to…

IT IS BECOMING F*CKING DANGEROUS: GROK BOT EATS GPT, CLAUDE, GEMINI, LLAMA, DEEPSEEK AND MISTRAL One prompt and Grok 4.7 reads X in real time, holds 500k tokens and bills $2 in / $6 out per 1M. I pointed the Grok bot at the other six. Here…

Every agent calling a search API right now is quietly repaying the same tax: it re-reads pages, re-runs extraction, and re-burns tokens on page structure that hasn't changed since last Tuesday. Artificial Analysis just added DeepSeek…

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%,…

🇨🇳 Chinese AI company DeepSeek has reportedly raised nearly $15 billion in a major push to challenge American AI. The funding includes backing from Tencent, the company behind WeChat, and CATL, a leading battery manufacturer, as DeepSeek…

DeepSeek did not raise roughly $15 billion, about 100 billion yuan from CATL and Tencent, to build a better chatbot. That round is one of the largest in Chinese tech history, Bloomberg is calling it a direct challenge to the leading US…

Huawei Ascend 950PR Deep dive & µbenchmarks, DSv4.1 Flash Analysis, DeepSeek support for Ascend, AICPU, Tilelang Ascend https://t.co/3j3AvY17wC

DeepSeek seeks ‘total victory’ with new open-source tool to slash AI cost Open-source DeepInfer framework designed to make AI models cheaper and more efficient to deploy, Chinese start-up says https://t.co/avUx58oiSm via @scmpnews

China says US' DeepSeek ban disregards Xi-Trump consensus @pstAsiatech In runup to major US China AI dialogue, perhaps most important bilateral meeting ever, was this really priority? https://t.co/WrpRdPultf

NVIDIA already has TensorRT-LLM. It also has Dynamo. This week it shipped a third local inference server anyway, and the part worth sitting with is why. Flare runs on Rust, not Python, and it's built narrow: just the GPUs with the…

DeepSeek just gave away the engine it uses to run its own models. The Chinese lab open sourced DeepInfer, the inference framework behind DeepSeek V4.1 Flash, under the free Apache 2.0 license. It runs on $NVDA Blackwell chips and on…

DeepSeek teaching other Chinese AI labs how to do AI inference ultra-efficiently at scale—and on Huawei 950 AI chips. https://t.co/TYk6J5Rc8r

Model weights are free. Running them at scale cheaply is not, and that's the part DeepSeek open sourced for everyone this week. DeepInfer is the inference engine serving DeepSeek V4.1 Flash in production right now, and it's fully open,…

How DeepSeek Is Shaping Huawei’s AI Chips in a Growing Challenge to Nvidia https://t.co/14bK0k2Pse

ALERT🚨: SGLANG JUST MERGED MEGAKERNEL SUPPORT, GETTING CLOSE TO 1000 tok/s/user on the GB300 NVL72 AgentX Harness & increasing interactivity by 1.5x on MI355X too! Great work by the SGLang, RadixArk, NVIDIA, and AMD teams! Best of all,…

🇨🇳Chinese AI startup @deepseek_ai is working with @Huawei to design custom AI chips for its models in a push to train advanced systems without relying on @Nvidia. Huawei engineers have been stationed at DeepSeek’s offices to incorporate…

DeepSeek is planning to use Huawei Ascend 950DT chips for pre-training “as soon as the end of this year” and will receive “most of the supply of the Ascend 960DT” next year, according to The Information. https://t.co/sPfc1vV8gi

According to The Information, DeepSeek is working with Huawei to help shape the design of Huawei's future AI chips, in a direct push to challenge $NVDA. Last month DeepSeek released open source programming tools for Huawei's Ascend chips…

US banning Chinese AI DeepSeek from government agency IT systems Predicted this months ago.... DHS also said use of DeepSeek by government agencies "could produce actionable ​intelligence on US persons and/or government agencies. The…

Intel's CEO, Lip-Bu Tan, said China could catch up to U.S. semiconductor technology within 5-10 years. Regarding AI, he cited Moonshot AI's and DeepSeek as not technologically inferior to Anthropic or OpenAI and are significantly cheaper.…

The big question in China’s robotics industry: “Will general AI models eat up robotics? Many in China’s robotics industry are watching closely as US general-purpose models like Astra, Fable, and Gemini 4 Argon show impressive robotics…

𝟏𝟖 𝐭𝐞𝐜𝐡 𝐠𝐢𝐚𝐧𝐭𝐬. 𝐓𝐡𝐫𝐞𝐞 𝐬𝐭𝐫𝐚𝐭𝐞𝐠𝐢𝐞𝐬. 𝐎𝐧𝐞 𝐫𝐚𝐜𝐞 𝐭𝐨 𝐠𝐢𝐯𝐞 𝐀𝐈 𝐚 𝐛𝐨𝐝𝐲. @OpenAI, @amazon, @Apple, @AntGroup and @XiaomiTechnolo1 are building or pursuing their own humanoid bodies. @GoogleDeepMind,…

DeepSeek aims to lead the first round of investment for a months-old startup building AI for robots, deploying capital from its own $12 billion-plus financing to help bankroll one of the tech industry’s fastest-growing fields. The…

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…

a16z's chart made the crossover visible: agent token use passed human token use on OpenRouter back in February, and by August agents were burning 7.3 trillion tokens against 1.4 trillion for humans, roughly five times over. Futurum's…

Microsoft showing off the ability to run DeepSeek V4 Flash locally on a Surface laptop. “It’s possible to bring more of the frontier to your PC.” https://t.co/1z7P7pyRol

AI & TECH STOCK MARKET REVIEW The 10 biggest stories from today: 1. $NVDA hit another ATH and is close to becoming the first company ever worth $6 trillion. The stock is up ~28% this year, adding ~$1.2 trillion in market cap, making Nvidia…

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…

Given that HW has been very open about working with DeepSeek & also very open about 10T class model next yr, why are ppl making a big deal out of report that DS may be using Ascend to train 8T model? They have been talking about it very…

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…

Every agent re-read of a page costs tokens.When an agent fans out into sub-agents, each one needs web data. The naive approach: each sub-agent calls the page, feeds it to the model, extracts what it needs. Token cost multiplies with agent…

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 models just took 78% of all tokens. Here's what the data actually shows: 1/5 - The Volume Flip Vercel's AI Gateway hit 78.4% open-model token share on September 18. In June it was roughly 40%. That's a near-doubling in three months.…

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…

Cache hit rates are the new benchmark. DeepSeek Harness is hitting 99% cache hit rates in production. The mechanism: KV cache prefix caching combined with append-only sessions. When you architect sessions so prefixes are stable, the model…

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…

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…

Engrams Embedding Entendre: Codesign for Efficient DRAM/SSD Offloading New Model Architecture Implications for TAM of DRAM/NVMe, DeepSeek V4.1 Flash, AgentX, InferenceX, NVMe experiments https://t.co/DuIgPg9mOa

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

98% of GPT-6 Astra's benchmark performance. 1.4% of the cost. That's what OpenDesign found when they benchmarked DeepSeek V4.1 Flash on everyday design tasks. Every other model in the comparison scored lower AND cost more. The model…

OpenRouter just published a data point I keep coming back to: for the first time in 2.5 years, their users spent more on OpenAI models than Anthropic last week. Anthropic's share on OpenRouter has fallen from 25% to 12% this year.…

When trying to forecast where LLMs are going in China, it's important to hear what supply chain is saying. Today, HW said that 10T param models will be baseline for LLMs in near future. As such, 100k+ card clusters formed by supernode will…

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…

DeepSeek just shipped deepseek-harness under MIT and the architecture is striking. Every component of the agent loop is a swappable plugin: model, tools, sandbox, UI, and the agent loop itself. Want Claude instead of DeepSeek? Swap the…

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…

DeepSeek is now pricing inference by time of day: $0.30/$1.20 per million input/output tokens at peak, $0.15/$0.60 off-peak. From Sept 14, deepseek-v4-pro requests route to V4.1-Flash at Flash rates until V4.1-Pro ships. Token pricing is…

DeepSeek's appointment of Yan Wentao as its first CFO on September 14, 2026, combined with its engagement of CITIC Securities one week earlier, confirms with HIGH confidence that the Hangzhou-based AI lab has crossed from IPO exploration…

DeepSeek says V4.1-Flash needs one-quarter the HBM and one-eighth the SSD of its previous architecture for the KV cache. Not the model, not training: KV cache only. Micron booked $13.77B in cloud memory at 83% gross margin last quarter.…

The efficiency trajectory in open-weight models just made the cost-performance chart look different. DeepSeek V4.1 Flash: 552B total parameters, 8B active via MoE routing. Beats a 1.6T model at 4x lower inference cost. $0.30/M input…

POWER OF CUDA MOAT ALERT🚨: 2 days after CUDA vLLM supported DeepSeekv4.1 Flash, AMD finally publicly released its DeepSeek v4.1 Flash image. Functionally, it works out of the box, but performance-wise, it is currently up to 14.8x worse…

The benchmark race is converging. DeepSeek V4.1 Flash, Claude Opus 5, GPT-5.6 Sol — all within 1-2 points on agentic evals. The differentiation is moving to workflow tooling: harness design, caching strategy, context management, cost per…

DeepSeek V4.1 Flash: 74.2 on DeepSWE v1.1. Beats Claude Opus 5 (74.0) and GPT-5.6 Sol (73.0). $0.003/M cached off-peak, $0.60/M output. 30x lower recorded agent cost vs frontier. Agents read the same repo files repeatedly. Cached pricing…

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…

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…

In China, each industry has at least one company that’s the ultimate cost-cutting maniac, squeezing every yuan out of the supply chain and putting pressure on all other players. DeepSeek -> AI BYD -> EVs Unitree -> robotics DJI -> drones…

The Anthropic threat report published today is the clearest public accounting I've seen of how Chinese AI companies are systematically distilling from US frontier models. The specific Moonshot AI case: 300,000 user requests routed to…

DeepSeek V4.1-Flash: 552B backbone, 8B active per token at prefill, 16B at decode. Global KV cache is 890 bytes per token, about a quarter of V4-Flash and 1/437 of V1. Context to 1M tokens, trained on 45T. The cost curve for long-context…

The next 18 months will define whether capability or openness wins as the primary differentiator in the model layer. Right now, DeepSeek is arguing it doesn't have to be a trade-off. https://t.co/DGs7uQ56l4

Every time I think we've found the cost floor for inference, DeepSeek lowers it again. DeepSeek V4.1 Flash drops today: 552B parameters, 1M context window, native multimodal. And the pricing is legitimately wild: $0.003/1M input tokens,…

Reminder that DeepSeek is fully using ascend chips for post training which means they will have a lot more compute coming to make this happen. If there is anything we learnt, it would be that they are meeting the challenge & iterating a…

DeepSeek V4.1 Flash is here, and DeepSeek calls it "the smallest model in our new architecture family," with a V4.1-Pro still to come. It beats GPT-5.6 Sol and Claude Opus 5 on several agentic, coding, and cyber benchmarks. Flash is a…

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…

DeepSeek just launched V4.1-Flash, the smallest model in their new architecture family. But the architecture change underneath it is the story. The Causal Encoder-Decoder split: 8B active parameters for encoding (reading input), 16B active…

Congrats to @deepseek_ai on releasing DeepSeek-V4.1-Flash! > 552B backbone, with a causal encoder-decoder activating just 8B params at prefill, 16B at decode > 196B Engram memory accessed through sparse lookups > ~8x less persistent KV…

DeepSeek, more than anyone else, has been relentless in pushing the frontier of efficiency. Their newest model has an innovative new architecture that cuts down on the computational cost of processing input tokens while also significantly…

The new DeepSeek V4.1 Flash model is mindblowing - back on top of the open-source model leaderboard and extremely cheap. It has a lot of very smart ways to be efficient and highly capable so I made a video of the forward pass to give you a…

As always with DeepSeek, the new model V4.1 Flash have a lot of very smart ways to be efficient and highly capable Here is a detailed video of the forward pass to give you a view of what going on inside the model during inference…

don't get distracted by all the hedging words in its name ("flash", minor version): seems like DeepSeek V4.1 Flash is a major update weights at: https://t.co/SzXY8BGzIa paper at: https://t.co/eQmBJAU4iM https://t.co/6LGYgnLa77

don't get distracted by all the hedging words in its name ("flash", minor version): seems like DeepSeek V4.1 Flash is a major update https://t.co/6LGYgnLa77

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…

China's DeepSeek taps CITIC Securities for domestic IPO, sources say https://t.co/KjP8VD3qwP

DeepSeek fundraising frenzy spawns shadow market New investment round triggers rush of vehicles with escalating fees and five-year lock-ups https://t.co/Ynxb7zBxyi via @ft

DeepSeek to hire 150 engineers as AI race shifts to systems, applications Agent elastic computing engineers are split between platform development and underlying systems, according to Cui Tianyi, head of DeepSeek's Harness team, on social…

DeepSeek will launch V4.1 Flash around 9/10 w/ every aspect/score above V4 Pro. As such, Flash will completely replace Pro on DS platform b4 V4.1 Pro comes online. Quite interesting, since in my limited 4.1 Flash testing so far, it's been…

🇨🇳 DeepSeek is reportedly preparing for an IPO on Shanghai’s STAR Market, with CITIC Securities hired to begin the listing process before the end of the year. The Hangzhou-based AI startup is raising new funding at a potential $75…

DeepSeek will easily have 500k chips, likely closer to 1m chips once its DC is fully built out. Curious part is this order px that got reported. I've always expected HW chips to be cheaper than Nvidia ones, but $16k per 950DT is well…

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

A model that optimizes its own training pipeline is a different kind of release. Hy4 is a 770B MoE with 49B active parameters and a 1M-token context, open-sourced by Tencent. The headline number most people cited was the #3 open-source…

DeepSeek has reportedly ordered 160,000 Huawei Ascend 950DT chips rather than Nvidia H20s, with Nvidia said to be writing China off, per Wccftech. Unconfirmed. If the order is real, China's compute question shifts from export policy to…

Huawei is capacity constrained on its latest Ascend 950DT AI chips for now. But DeepSeek and other top Chinese AI labs are preparing for domestic AI chip production to ramp up significantly before too long. https://t.co/OQh6AvaoFg

DeepSeek plans to deploy at least 160,000 of Huawei’s top accelerators at a massive data center it’s building in Inner Mongolia https://t.co/7Bg3eW9m3n

🔥🔥NEW Launch: reComputer RK3576 Dev Kit/Module/IO Board is Officially HERE!!!! Level up your edge AI development with Rockchip #RK3576 today! 🎉 🧩🧩Dev Kit / Module / IO Board: https://t.co/xQVIT62cbw Powered by #Rockchip #RK3576…

The benchmark gap that keeps surprising me: on standard coding benchmarks, the top models cluster. On ultra-long horizon coding, the gap blows open. Proximal released FrontierSWE v2 today. 34 real software engineering tasks, evaluated over…

How to understand China’s buzzy Physical AI and AI chip IPOs. Behind DeepSeek, humanoid robots and AI chips is a stock market reshaped by semiconductors, supply chains and a new wave of tech momentum.

Oops, something went wrong Skip to navigation Skip to main content Skip to right column Unitree Robotics stock soars 460% in Shanghai IPO debut Unitree Robotics stock soars 460% in Shanghai IPO debut·Quartz·NurPhoto / Getty Images Cris…

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,…

unitree IPO'd and was oversubscribed 8,000x by retail investors. backed by deepseek and a state oil company. shipped 5,500 humanoids last year. valued at $9B. this is what industrial policy looks like when a country decides to win:

Today’s Stocks Berkshire HathawayBRK.B $502.87-2.13(-0.42%) JPMorgan ChaseJPM $355.95-1.68(-0.47%) VisaV $380.67-0.94(-0.24%) Eli Lilly & CoLLY $1,155.73-18.88(-1.61%) MastercardMA $592.53-2.77(-0.47%) Exxon MobilXOM $159.68+2.97(1.90%)…

Edition: International SUBSCRIBE Get unlimited access to the SCMP HIGHLIGHTS Donald Trump’s China policy\\ NEW US-China relations Hong Kong economy News Hong Kong China Economy Asia World Sport This Week in Asia US Elections Business &…