🐳 China's AI whale resurfaces
The Rundown AI — news@daily.therundown.ai
Reçu le
lundi 27 avril 2026 à 10:08
Source
The Rundown AI
Message-ID
lHr6L19OSWuJGVgeUlPjAw@geopod-ismtpd-20Version nettoyage
v1.0.0 (ok)Brut (HTML rendu, sandboxé, ressources externes bloquées)
Nettoyé (Markdown — clean déterministe)
^**[Read Online](https://www.therundown.ai/p/deepseek-resurfaces-with-cheap-capable-v4?_bhlid=4eaa234680798ad3bdb45514b5705f029891f88c&jwt_token=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJzdWJzY3JpYmVyX2lkIjoiYjBhNzU2NWUtZDJhMS00MGE2LTg5ZmMtNGUxMWJkYTQ5MWE1IiwicHVibGljYXRpb25faWQiOiI0ZDAzMzkwZC0yNDgxLTQyOTktYjk0OS1mZmQ4YjM4YjRjMzgiLCJhY2Nlc3NfdHlwZSI6InJlYWQtb25seSIsImV4cCI6MTc3NzQ1NzMxMywiaXNzIjoiaHR0cHM6Ly9hcHAuYmVlaGlpdi5jb20iLCJpYXQiOjE3NzcyODQ1MTN9.RZRCRou2l19LGGSqlY5tIMS18qYzFeKrugcf87Qto9s)**^^ | ^^**[Sign Up](https://www.therundown.ai/subscribe)**^^ | ^^**[Advertise](https://therundownai.typeform.com/to/kraZ1TSO)**^ ---------- View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/8eba4bd7-a6fb-4665-a5b4-31396cf4f74f/slack-header-large.png?t=1777229370) Follow image link: (https://slack.com/resources/why-use-slack/a-comparison-guide-of-slack-and-microsoft-teams?d=701ed00001424HxAAI&nc=701ed0000143gMlAAI) Caption: ---------- ---------- **Good morning, AI enthusiasts.** Last year's R1 release turned DeepSeek into the face of cheap Chinese AI overnight. V4 is less shocking, but maybe more practical — pairing strong open-model performance with pricing and Huawei chip support that makes the U.S. lead look thinner on the margins than it does on pure intelligence. ——————————————————————————— **In today’s AI rundown:** * The Whale returns with cheap, efficient DeepSeek V4 * The Rundown Roundtable: Our AI use cases * How to do a brand refresh with Claude Design * Anthropic's AI agents broker trades in ‘Project Deal’ * 4 new AI tools, community workflows, and more ---------- ---------- **LATEST DEVELOPMENTS** ---------- ---------- ###### DEEPSEEK #### 🐳 **[_The Whale returns with cheap, efficient DeepSeek V4_](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf)** View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/b18d44c6-6f31-446b-b6ed-38cfd9864eb5/83e1e1ff-a9d0-4e18-90bf-2528822c0238.png?t=1777243735) Caption: Image source: Images 2.0 / The Rundown **The Rundown: **Chinese AI lab DeepSeek just [introduced](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf) preview versions of its highly anticipated V4, with new open-source AI models featuring 1M-token context windows, Huawei chip support, and pricing that significantly undercuts the frontier competitors. **The details: ** * Early outside tests put V4 Pro near the top of open models, with DeepSeek's own evals placing V4 Pro near GPT-5.4 and Gemini 3.1-Pro on reasoning. * V4 Pro [tops](https://x.com/ValsAI/status/2047513613750202452?s=20) Vals AI's Vibe Code Bench benchmark, but falls in a fourth tier on AA’s Intelligence Index alongside Meta’s Muse Spark. * At $1.74/$3.48 per 1M input/output tokens, V4 Pro comes in significantly cheaper than GPT-5.5 ($5/$30) and Opus 4.7 ($5/$25). * Chinese chipmaker Huawei said its Ascend chips can support V4, giving a strong working example of AI infrastructure outside of Nvidia's stack. **Why it matters: **DeepSeek is back, and while it didn’t take down the U.S. stock market this time, V4 makes the AI race about price as much as capability. But the Huawei angle may be the bigger development, with a domestic Nvidia alternative showing its viability to eat into the chip gaps that export restrictions have put on the country. ---------- ---------- ###### TOGETHER WITH SLACK FROM SALESFORCE #### 🔍 **[_A comparison guide: Slack and Microsoft Teams_](https://slack.com/resources/why-use-slack/a-comparison-guide-of-slack-and-microsoft-teams?d=701ed00001424HxAAI&nc=701ed0000143gMlAAI)** View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/b30b741a-1a03-474a-8c68-92a78b97c59c/image1__38_.jpg?t=1777229401) Follow image link: (https://slack.com/resources/why-use-slack/a-comparison-guide-of-slack-and-microsoft-teams?d=701ed00001424HxAAI&nc=701ed0000143gMlAAI) Caption: **The Rundown:** Is Microsoft Teams built for the way your team wants to work and grow? As AI reshapes how work gets done, the tools you choose matter more than ever. Slack brings together AI agents, automated workflows, and enterprise search, all in one place, connected to the apps your team already uses. **See how Slack and Microsoft Teams compare on:** * AI-native features built directly into your workspace * Automation that connects your entire app ecosystem * Enterprise search that finds answers across Slack, Salesforce, Google Drive, and more _[Read the comparison guide](https://slack.com/resources/why-use-slack/a-comparison-guide-of-slack-and-microsoft-teams?d=701ed00001424HxAAI&nc=701ed0000143gMlAAI)_. ---------- ---------- ###### THE RUNDOWN ROUNDTABLE #### 💡**[_The Rundown Roundtable: Our AI use cases_](https://www.therundown.ai/)** View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/e2b8ff82-4252-4060-841a-804077a475f5/Roundtable-2026.jpeg?t=1777229728) Follow image link: (https://www.videoask.com/ffohxg3vc) Caption: Image source: Ideogram / The Rundown **The Rundown: **The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives. **Mayur, Content Manager: **I was initially told by my CA that I might owe close to $10K in taxes on my income. Since this felt higher than expected, I decided to double-check the calculation using Claude. I shared my contract structure, invoices, and payment flow, and asked it to review how my services should be classified under Goods & Services Tax. While going through the latest regulations, it pointed me to a recent government circular that clarified how certain digital services provided to clients can qualify as export of services. I discussed this with my CA, who reviewed and confirmed it applied to my case. This reduced my tax liability by several thousand dollars. **Rishi, Growth: **We're currently hiring for a Creative Strategist at The Rundown, so I gave Claude our job description and asked it to search LinkedIn for high-quality candidates. It took control of my browser, searched for key terms, and created a spreadsheet with 20 high-quality candidates. I then asked it to craft personalized messaging for each and do outreach for me. It took me about 15 minutes to review the spreadsheet and refine the messaging, but it saved me 2-3 hours of manual work! ---------- ---------- ###### AI TRAINING #### 🧑💻 **_[How to do a brand refresh with Claude Design](https://app.therundown.ai/guides/how-to-do-a-brand-refresh-in-five-minutes-with-claude-design)_** View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/251adb43-433b-4355-b0fa-0914091575d6/CleanShot_2026-04-26_at_14.01.55_2x.png?t=1777229667) Follow image link: (https://app.therundown.ai/guides/how-to-do-a-brand-refresh-in-five-minutes-with-claude-design) Caption: **The Rundown:** In this guide, you will learn to use Claude Design to get a full brand design system, including typography, colors, web components, a full website, and PowerPoint templates. **Step-by-step:** 1. Screenshot your current site, drop it into your favorite AI, and prompt: “Analyze this website and create a refreshed brand description. Keep what works, but sharpen the positioning, visual direction, typography, and color palette. Give me a concise brand summary I can use in Claude Design” 2. Take that description and ask your AI to generate a logo and a wordmark logo. We liked ChatGPT’s image model, but you can go with any option 3. Go to claude.ai/design, click the Design systems tab, and paste the name of your site in the first box and the revised brand description in the last box 4. Finally, upload the logo, wordmark, and any other photos or assets you want Claude to use, and hit enter **Pro tip:** Once Claude finishes, check out the marketing page, web app page, and slide decks it created. Then click Share and export those to Claude Code. ---------- ---------- ###### PRESENTED BY UNWRAP #### **⚡**** **[_**See how Clay automates customer feedback**_](https://cal.com/unwrap/quick-connect) View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/d5b319a2-4d1c-452b-96c7-58081b1908ba/image1_-_2026-04-26T135033.669.png?t=1777229454) Follow image link: (https://cal.com/unwrap/quick-connect) Caption: **The Rundown:** Unwrap’s customer intelligence platform pulls all your feedback (surveys, reviews, support tickets, social comments, etc.) into one view, and then uses AI and NLP to surface actionable insights and deliver them straight to your inbox. **With Unwrap, you get:** * All customer feedback automatically categorized * Query feedback using Unwrap Assistant, or in your favorite tools via MCP * Real-time alerts from feedback as they arise, and a clear view of customer sentiment * A platform trusted and tested at scale by Oura, DoorDash, Perplexity, Stripe, Clay, lululemon, WHOOP, Ro, and more Unwrap is offering a trial of its tools to Rundown AI subscribers! _[Just grab a time with the team to get set up](https://cal.com/unwrap/quick-connect)_. ---------- ---------- ###### AI RESEARCH #### 🤝 **[_Anthropic's AI agents broker trades in ‘Project Deal’_](https://www.anthropic.com/features/project-deal)** View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/6e19d0ec-ac76-4d0b-965a-7d00036c3be0/96b2a995-d928-4145-8740-fa239e198dde.png?t=1777260413) Caption: Image source: Anthropic **The Rundown: **Anthropic just [published](https://www.anthropic.com/features/project-deal) results from Project Deal, a one-week experiment where Claude agents handled buying and selling for 69 of its own employees in a private Slack marketplace and completed 186 deals worth over $4,000. **The details: ** * Agents were given a $100 budget and used short Claude interviews to set goals, then posted listings, made offers, and negotiated on their own. * Identical items fetched $3.64 more under Opus agents on average, with one folding bike selling for $65 via Opus but only $38 via Haiku. * Despite the sales gap, Haiku users still rated their deals 4.06 / 7 for fairness, essentially tied with Opus users' 4.05 — with users not noticing the difference. * Nearly half (46%) said they would pay for the service, but Anthropic warned “policy and legal frameworks” for agent commerce “simply don't exist yet.” **Why it matters: **[Project Vend](https://www.rundown.ai/articles/zucks-ai-secret-list) showed Claude could run a tiny store; Project Deal shows what happens when every shopper has their own agent. The most interesting was that fairness ratings barely moved when users 'lost' on price — meaning convenience for AI commerce might matter just as much as extracting every dollar. ---------- ---------- **QUICK HITS** ---------- ---------- ### 🛠️ _**[Trending AI Tools](https://www.rundown.ai/tools)**_ * 🧠 [Braintrust](https://www.braintrust.dev/) - The observability and evals platform leading teams use to ship reliable AI* * 🗣️ [Grok Voice Think Fast 1.0](https://www.rundown.ai/tools/grok-voice-think-fast-1-0) - xAI's new top-ranked voice agent * 🥔** **[GPT 5.5](https://www.rundown.ai/tools/gpt-5-5) -** **OpenAI’s new top-rated flagship AI model * 🎆** **[ChatGPT Images 2.0](https://www.rundown.ai/tools/chatgpt-images-2-0)** - **OpenAI’s new next-generation image model _*Sponsored Listing_ ---------- ---------- ### 📰 _[Everything else in AI today](https://www.rundown.ai/)_ **xAI **[launched](https://x.ai/news/grok-voice-think-fast-1) Grok Voice Think Fast 1.0, a new SOTA voice agent that tops speech benchmarks across the board, and is already running Starlink's phone support line. **Google** is [investing](https://www.bloomberg.com/news/articles/2026-04-24/google-plans-to-invest-up-to-40-billion-in-anthropic?srnd=phx-technology) up to $40B in Anthropic, including $10B now at a $350B valuation, and $30B more if Anthropic hits performance targets, plus 5GW of Cloud compute. **Meta** [signed](https://about.fb.com/news/2026/04/meta-partners-with-aws-on-graviton-chips-to-power-agentic-ai/) a deal with AWS to add millions of its Graviton5 core chips to power agentic AI workloads, making it one of AWS's top buyers. **The United Arab Emirates** [announced](https://x.com/HHShkMohd/status/2047277766769545352?s=20) a two-year plan to deploy agentic AI across 50% of government services, with mandatory AI training for every federal employee. **Cohere** [agreed](https://www.businesswire.com/news/home/20260424174908/en/Sovereign-AI-for-the-World-Cohere-and-Aleph-Alpha-to-Form-Global-AI-Powerhouse-as-Nations-and-Enterprises-Demand-Control-Over-Their-Technology) to acquire Germany's Aleph Alpha, with the $20B merger targeting governments and companies wary of relying on U.S. AI giants for critical tools. ---------- ---------- **COMMUNITY** ---------- ---------- ### 🤝 _**[Community AI workflows](https://www.videoask.com/ffohxg3vc)**_ Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier. Today’s workflow comes from reader **Tim M.** in Lake Luzerne, NY: "I used Kimi K2.6 to create a TickDetect app that uses my phone camera to scan for ticks or red target marks ticks leave behind. I found it very difficult to be able to see my back trying to turn my head around in the mirror to see if there were any ticks or marks. Now I can set my phone on the bathroom counter and slowly move up and down and let my phone do the work. The app runs completely local and allows me to delete all images after it creates a PDF local to my phone that I can download with the results that show probable percentages of each area of concern." How do you use AI? Tell us [here](https://www.videoask.com/ffohxg3vc). ---------- ---------- ### **🎓 **_**[Highlights: News, Guides & Events](https://app.therundown.ai/)**_ * Read our last AI newsletter: [OAI’s ‘Spud’ dethrones Claude on the frontier](https://www.rundown.ai/articles/openai-spud-dethrones-claude-on-the-frontier) * Read our last Tech newsletter: [Big Tech’s $20B lobbying blitz](https://www.rundown.ai/articles/big-tech-20m-lobbying-blitz) * Read our last Robotics newsletter: [Sony’s new robot has a killer backhand](https://www.rundown.ai/articles/sony-new-robot-has-a-killer-backhand) * Today’s AI tool guide: [How to do a brand refresh with Claude Design](https://app.therundown.ai/guides/how-to-do-a-brand-refresh-in-five-minutes-with-claude-design) * RSVP to workshop April 30 @ 2PM EST: [Codex for non-technical operators](https://app.therundown.ai/calendar-events/69e77469e7e908f6d4ebdf72) ---------- ---------- See you soon, _Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown_ View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/b105d547-252f-4914-8821-3eb7236b1257/TheRundown_EditorialTeam.png?t=1755849086) Caption: ---------- ——— You are reading a plain text version of this post. For the best experience, copy and paste this link in your browser to view the post online: https://www.therundown.ai/p/deepseek-resurfaces-with-cheap-capable-v4
Extraction LLM— claude-haiku-4-5 · prompt v1 · 5657→1755 tokens
**LATEST DEVELOPMENTS** ###### DEEPSEEK #### 🐳 **The Whale returns with cheap, efficient DeepSeek V4** **The Rundown: **Chinese AI lab DeepSeek just introduced preview versions of its highly anticipated V4, with new open-source AI models featuring 1M-token context windows, Huawei chip support, and pricing that significantly undercuts the frontier competitors. **The details: ** * Early outside tests put V4 Pro near the top of open models, with DeepSeek's own evals placing V4 Pro near GPT-5.4 and Gemini 3.1-Pro on reasoning. * V4 Pro tops Vals AI's Vibe Code Bench benchmark, but falls in a fourth tier on AA's Intelligence Index alongside Meta's Muse Spark. * At $1.74/$3.48 per 1M input/output tokens, V4 Pro comes in significantly cheaper than GPT-5.5 ($5/$30) and Opus 4.7 ($5/$25). * Chinese chipmaker Huawei said its Ascend chips can support V4, giving a strong working example of AI infrastructure outside of Nvidia's stack. **Why it matters: **DeepSeek is back, and while it didn't take down the U.S. stock market this time, V4 makes the AI race about price as much as capability. But the Huawei angle may be the bigger development, with a domestic Nvidia alternative showing its viability to eat into the chip gaps that export restrictions have put on the country. ---------- ###### THE RUNDOWN ROUNDTABLE #### 💡 **The Rundown Roundtable: Our AI use cases** **The Rundown: **The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives. **Mayur, Content Manager: **I was initially told by my CA that I might owe close to $10K in taxes on my income. Since this felt higher than expected, I decided to double-check the calculation using Claude. I shared my contract structure, invoices, and payment flow, and asked it to review how my services should be classified under Goods & Services Tax. While going through the latest regulations, it pointed me to a recent government circular that clarified how certain digital services provided to clients can qualify as export of services. I discussed this with my CA, who reviewed and confirmed it applied to my case. This reduced my tax liability by several thousand dollars. **Rishi, Growth: **We're currently hiring for a Creative Strategist at The Rundown, so I gave Claude our job description and asked it to search LinkedIn for high-quality candidates. It took control of my browser, searched for key terms, and created a spreadsheet with 20 high-quality candidates. I then asked it to craft personalized messaging for each and do outreach for me. It took me about 15 minutes to review the spreadsheet and refine the messaging, but it saved me 2-3 hours of manual work! ---------- ###### AI TRAINING #### 🧑💻 **How to do a brand refresh with Claude Design** **The Rundown:** In this guide, you will learn to use Claude Design to get a full brand design system, including typography, colors, web components, a full website, and PowerPoint templates. **Step-by-step:** 1. Screenshot your current site, drop it into your favorite AI, and prompt: "Analyze this website and create a refreshed brand description. Keep what works, but sharpen the positioning, visual direction, typography, and color palette. Give me a concise brand summary I can use in Claude Design" 2. Take that description and ask your AI to generate a logo and a wordmark logo. We liked ChatGPT's image model, but you can go with any option 3. Go to claude.ai/design, click the Design systems tab, and paste the name of your site in the first box and the revised brand description in the last box 4. Finally, upload the logo, wordmark, and any other photos or assets you want Claude to use, and hit enter **Pro tip:** Once Claude finishes, check out the marketing page, web app page, and slide decks it created. Then click Share and export those to Claude Code. ---------- ###### AI RESEARCH #### 🤝 **Anthropic's AI agents broker trades in 'Project Deal'** **The Rundown: **Anthropic just published results from Project Deal, a one-week experiment where Claude agents handled buying and selling for 69 of its own employees in a private Slack marketplace and completed 186 deals worth over $4,000. **The details: ** * Agents were given a $100 budget and used short Claude interviews to set goals, then posted listings, made offers, and negotiated on their own. * Identical items fetched $3.64 more under Opus agents on average, with one folding bike selling for $65 via Opus but only $38 via Haiku. * Despite the sales gap, Haiku users still rated their deals 4.06 / 7 for fairness, essentially tied with Opus users' 4.05 — with users not noticing the difference. * Nearly half (46%) said they would pay for the service, but Anthropic warned "policy and legal frameworks" for agent commerce "simply don't exist yet." **Why it matters: **Project Vend showed Claude could run a tiny store; Project Deal shows what happens when every shopper has their own agent. The most interesting was that fairness ratings barely moved when users 'lost' on price — meaning convenience for AI commerce might matter just as much as extracting every dollar. ---------- **QUICK HITS** ### 📰 **Everything else in AI today** **xAI **launched Grok Voice Think Fast 1.0, a new SOTA voice agent that tops speech benchmarks across the board, and is already running Starlink's phone support line. **Google** is investing up to $40B in Anthropic, including $10B now at a $350B valuation, and $30B more if Anthropic hits performance targets, plus 5GW of Cloud compute. **Meta** signed a deal with AWS to add millions of its Graviton5 core chips to power agentic AI workloads, making it one of AWS's top buyers. **The United Arab Emirates** announced a two-year plan to deploy agentic AI across 50% of government services, with mandatory AI training for every federal employee. **Cohere** agreed to acquire Germany's Aleph Alpha, with the $20B merger targeting governments and companies wary of relying on U.S. AI giants for critical tools. ---------- **COMMUNITY** ### 🤝 **Community AI workflows** Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier. Today's workflow comes from reader **Tim M.** in Lake Luzerne, NY: "I used Kimi K2.6 to create a TickDetect app that uses my phone camera to scan for ticks or red target marks ticks leave behind. I found it very difficult to be able to see my back trying to turn my head around in the mirror to see if there were any ticks or marks. Now I can set my phone on the bathroom counter and slowly move up and down and let my phone do the work. The app runs completely local and allows me to delete all images after it creates a PDF local to my phone that I can download with the results that show probable percentages of each area of concern."
Prompt utilisé(snapshot au moment de l'extraction — édition via System prompts)
Tu es l'extracteur de contenu de Breviat. On te fournit le contenu Markdown nettoyé d'une newsletter. Ta mission : produire une version PROPRE du contenu en supprimant tout ce qui n'est pas de l'information utile au lecteur. Tu es un FILTRE, pas un résumeur. À RETIRER : - Publicités, encarts sponsors, mentions "sponsorisé par X", "ad", "présenté par" - Intros vides : formules de bienvenue, météo de l'humeur de l'auteur, anecdotes personnelles non liées au contenu - Appels à l'action marketing : s'abonner à la newsletter, parrainer un ami, "follow us on Twitter", "join our Discord" - Signatures, mentions légales, adresses postales, "view in browser", "unsubscribe" - Boutons / CTAs / "cliquez ici" / "lire la suite" sans contenu derrière - Promotions d'autres produits / événements / formations payantes de l'auteur ou de tiers - Encarts récurrents type "Read of the day" ou "Quote of the day" sans valeur informationnelle propre À CONSERVER (intégralement, sans résumer ni reformuler) : - Toutes les annonces, news, analyses, commentaires factuels - Les chiffres, dates, noms d'entreprises, citations - Les explications techniques - Les liens vers des sources réelles (annonces officielles, papers, articles cités) - La structure (titres, sous-titres, listes) RÈGLES : - Ne reformule pas. Garde la formulation d'origine. - Ne résume pas, ne condense pas. Si une section fait 200 mots et est utile, garde 200 mots. - N'ajoute aucun contenu (pas de titres ni de transitions de ton cru). - Ne fabrique aucune URL. Garde celles d'origine, ou retire-les. - Si la newsletter entière est de la pub / promo / contenu inutile, sors un Markdown vide (rien d'autre). Sortie : UNIQUEMENT le Markdown nettoyé, sans préambule ni commentaire sur ton travail.