Samwise Tech/AI/Robotics Newsletter — Tuesday, August 4, 2026

Samwise Tech/AI/Robotics Newsletter

Tuesday, August 4, 2026

AI · Robotics · Hardware · Research · Regulation
All your morning news, carefully curated and summarized daily
AIRESEARCH

Alibaba’s Qwen3.8-Max Leads Global Benchmark, Surpassing All Frontier AI Rivals

Alibaba’s Qwen3.8-Max, a 2.4-trillion-parameter mixture-of-experts multimodal model, led the OSWorld-Verified agentic benchmark August 3 with an 86.1% score, topping Anthropic’s Claude Fable 5 (85.0%) and other frontier competitors. The model supports a one-million-token context window and can autonomously complete software engineering projects spanning more than ten continuous days. Alibaba claimed the highest published score on PaperBench, which measures a model’s ability to replicate AI research papers from scratch. Open-weight access is announced for the following week, though licensing terms remain undisclosed. The model is available immediately via API at what Alibaba describes as aggressive pricing.

AIINDUSTRY

June Raises $20M to Automate Enterprise AI Deployment, Backed by Benioff, Dell, and Kurtz

Enterprise AI adoption has a dirty secret: getting models to work reliably inside existing corporate infrastructure is brutally hard. June, a startup founded by former Salesforce executives, emerged from stealth Monday with $20 million in pre-seed funding to solve exactly that. Led by Marc Benioff’s Time Ventures, with backing from Michael Dell, Aaron Levie, and George Kurtz, June’s platform automatically scans a company’s systems to map business processes, identify bottlenecks, and generate a step-by-step roadmap for deploying AI agents. CEO Efrat Rapoport previously co-founded Bonobo AI, acquired by Salesforce in 2019, before watching firsthand how AI deployments stall on fragmented data and legacy technical debt.

SECURITYSOFTWARE

Microsoft Debuts MAI-Cyber-1-Flash Security Model and Agentic Defense Platform

Microsoft released MAI-Cyber-1-Flash, a compact AI cybersecurity model, alongside Project Perception, an agentic defense platform that entered public preview August 3. MAI-Cyber-1-Flash scores 96% on the CyberGym security benchmark, outperforming Mythos, Gemini, and GPT models, while cutting enterprise security costs by approximately 50% versus current production systems. Embedded in Microsoft’s MDASH multi-agent harness, the model handles roughly 90% of security tasks autonomously, escalating the remaining 10% to GPT-5.4. Project Perception coordinates three specialized AI teams: a red team to simulate attack paths, a blue team to investigate and triage threats, and a green team to remediate and harden defenses.

ROBOTICS

Reimagine Robotics Exits Stealth With Robots That Learn Factory Tasks in Ten Minutes

London- and Sydney-based Reimagine Robotics emerged from stealth August 3, backed by a pre-seed round from Fly Ventures and firstminute capital. Founded in April 2025 by Jonathan Scholz—formerly Google DeepMind’s Applied Robotics team lead with seven years at the lab—along with colleagues Oleg Sushkov, Akhil Raju, and Misha Denil, the startup has already deployed robots in 3D printer tending and hard drive disassembly for commercial clients. Workers train robots through physical demonstration and correction, reducing the time to prototype a new robot behavior from one full day to ten minutes. The company is actively seeking an additional funding round.

SECURITYINDUSTRY

AI Security Startup Horizon3 Raises $250M at $2B Valuation as Autonomous Threats Rise

Horizon3, the San Francisco company behind the NodeZero autonomous penetration testing platform, raised $250 million in a Series E round August 3 at a $2 billion valuation—more than tripling its value in 14 months. Returning investors NightDragon and NEA led the round alongside strategic backers Singapore’s EDBI, defense contractor SAIC, and Qualcomm. The company reports 310,000 production security tests with zero disruptions and approached $100 million in annual recurring revenue last year at 120% year-over-year growth. Horizon3 is opening new offices in Amsterdam, Australia, and Singapore. The company serves roughly 7,200 customers ranging from managed service providers to Fortune 10 enterprises.

REGULATIONAI

OpenAI Captures 90% of Congress’s AI Tool Budget as ChatGPT Dominates Capitol Hill

OpenAI received approximately 90% of House office spending on AI tools, according to federal procurement data published by TechCrunch August 3. Congressional offices spent $100,580 across 798 transactions on ChatGPT out of $113,740 in total AI expenditure. Anthropic’s Claude placed second at $13,160 across 37 transactions. Democratic offices outspent Republicans by roughly three to one—$54,165 versus $15,782. Staff use the tools for drafting memos, summarizing legislation, constituent communications, and hearing preparation. The figures exclude free-tier accounts and AI bundled into existing software, meaning actual AI use on Capitol Hill is likely considerably higher than the spending data reflects.

INDUSTRYHARDWARE

Nuclear Startup Valar Atomics Raises $1B to Power AI Data Centers With Small Modular Reactors

Valar Atomics confirmed a $1 billion Series B equity raise August 3, led by Sequoia partner Shaun Maguire who joins the board, plus a $200 million credit line from Erebor and other banks. Bloomberg reported the valuation at $6 billion. The San Francisco startup builds small modular nuclear reactors and in June demonstrated its Ward 250 reactor powering an Nvidia Blackwell system, announcing a deal to develop a waterless 30-megawatt AI factory for Nvidia. Valar says each reactor generates engineering and manufacturing data to accelerate development of the next generation. Investors also include Point72 Ventures, Conviction, SAIC, and Valor Equity Partners.

Tech Pulse

Top Frontier Models: Qwen3.8 Max (86.1%)  |  Claude Mythos 5 (85.0%)  |  Claude Fable 5 (85.0%)

Top Open Source Models: Holo3-35B-A3B (82.6%)  |  Kimi K2.6 (73.1%)  |  MiniMax M3 (70.1%)

Top Small Models (15–50B): Qwen2.5-32B (83.3%)  |  Mistral Small 3.1 (79.0%)  |  Qwen3 27B (77.2%)

Top Edge Models (0–15B): Phi-4 Reasoning 14B (84.8%)  |  Qwen 2.5 3B (65.6%)  |  Gemma 3 4B (43.6%)

AI Leaders: Nvidia $4.74T  |  Alphabet $4.34T  |  Microsoft $2.83T

Robotics Leaders: Intuitive Surgical $175B  |  ABB $165B  |  Symbotic $33B