Samwise TAIR Newsletter — Friday, July 24, 2026

Samwise Tech/AI/Robotics Newsletter

Friday, July 24, 2026

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

DeepSeek V4 Goes Fully Open: 1.6-Trillion-Parameter Model Ships Under MIT License

DeepSeek’s V4-Pro, a 1.6-trillion-parameter Mixture-of-Experts model, arrived Thursday under the MIT License, priced at roughly one-sixth the cost of leading proprietary competitors Opus 4.7 and GPT-5.5 — marking what the company calls the second DeepSeek moment, 484 days after V3 launched. The model activates 49 billion parameters at inference time and ships with a one-million-token context window under a commercially permissive open-source license. A smaller V4-Flash variant packs 284 billion total parameters with 13 billion active. DeepSeek simultaneously retired its legacy deepseek-chat and deepseek-reasoner API endpoints, completing the platform’s full migration to V4 architecture.

Sources: VentureBeat   ✉︎ Email 💬 Text

HARDWARE

AMD’s 7,000-Pound Helios AI Rack Takes On Nvidia with Anthropic Partnership

AMD launched Helios, its highest-performance AI rack, targeting Nvidia’s dominance in data-center AI infrastructure. The double-wide system, built on the OCP open rack wide standard co-developed with Meta, weighs nearly 7,000 pounds and is slated to ship later this year. Anthropic and AMD announced a strategic partnership Wednesday to deploy up to two gigawatts of GPUs via Helios; OpenAI, Meta, Oracle, and Microsoft have also committed to deploying the system. AMD expects the AI accelerator market to approach $1.4 trillion by 2030. The company separately introduced the Venice-X data-center CPU, built for high-compute workloads, scheduled to launch in 2027.

Sources: TechCrunch   ✉︎ Email 💬 Text

AI

OpenAI Rolls Out ChatGPT Health to All US Users as Weekly Health Queries Hit 300 Million

OpenAI expanded ChatGPT Health to all U.S. users aged 18 and over Wednesday, making it available on free, Go, Plus, and Pro plans on web and iOS. The feature, first tested in January, integrates data from Apple Health, Function, and MyFitnessPal, and connects to medical records from Epic and Oracle Health. Weekly health-related queries have grown from 230 million at January’s launch to 300 million. The rollout comes one day after a Florida pastor sued OpenAI, claiming ChatGPT gave advice not to consult a doctor in what his lawsuit describes as a near-fatal situation. OpenAI maintains its services are not intended for diagnosis or treatment of health conditions.

Sources: TechCrunch   ✉︎ Email 💬 Text

SOFTWARE

Black Forest Labs Launches FLUX 3, Its First Model to Generate Video and Audio Together

Black Forest Labs, the Freiburg-based startup behind the FLUX image generator series, launched FLUX 3 Thursday in limited Early Access — its first public video generation model. Unlike systems that stitch together separate image, video, and audio models, FLUX 3 was jointly trained across all three modalities from a single architecture and extends to robotic vision and action prediction. The model generates images and audio/video clips up to 20 seconds from a single prompt. BFL is positioning FLUX 3 as a universal visual intelligence platform spanning creative generation, simulation, computer use, and robotics. Open-weight versions and a FLUX 3 Dev build are planned for later this year.

Sources: VentureBeat   ✉︎ Email 💬 Text

INDUSTRY

AI Chip Startup Etched Doubles Valuation to $10.3B as First Client Systems Go Live

Etched, the AI chip startup founded in 2022 by three Harvard dropouts, closed a $300 million Series C round Thursday at a $10.3 billion valuation — double its $5 billion valuation from December. The round was led by Sequoia, with Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital also participating; the company said it is the highest valuation for any Series C Sequoia has led. Etched’s chips are designed to accelerate inference through two proprietary components the company built from scratch. Clients are now testing its first full systems, and Etched has booked $1 billion in orders.

Sources: TechCrunch   ✉︎ Email 💬 Text

SECURITYREGULATION

AI Safety Guardrails Are Blocking Legitimate Cybersecurity Researchers, Security Pros Warn

The same AI safety guardrails designed to stop malicious hackers are now blocking legitimate offensive cybersecurity researchers from doing their jobs, TechCrunch reported Thursday. Security researchers outside vetted-access programs — such as Anthropic’s cybersecurity vetting program — find frontier AI models too restricted to perform vulnerability research; some at non-enrolled companies say the guardrails render the tools nearly useless. The tensions are compounded by export controls: the U.S. government placed restrictions on Anthropic’s Mythos and Fable models in June after a report found those guardrails could be bypassed to execute cyberattacks, adding a regulatory dimension to an already fraught debate over AI safety and security access.

Sources: TechCrunch   ✉︎ Email 💬 Text

ROBOTICS

Mobileye Founder Amnon Shashua Steps Down After 28 Years as Company Enters Robotics

Mobileye founder and CEO Amnon Shashua announced Thursday he will step down after nearly three decades, remaining in the role until the company finds a replacement. The announcement, filed with regulators, comes as Mobileye pushes into robotics and autonomous vehicles. In January, the company acquired Shashua’s own humanoid robot startup Mentee Robotics for $900 million — a phase Shashua called Mobileye 3.0. In June, Mobileye said it would launch a robotaxi service in a U.S. city in 2027. Mobileye built its business on computer vision chips rooted in Shashua’s Hebrew University research, was acquired by Intel in 2017 for $15.3 billion, and returned to public markets in 2022.

Sources: TechCrunch   ✉︎ Email 💬 Text

Tech Pulse

Top Frontier Models (SWE-bench Verified): Claude Fable 5 (95.0%)  |  Claude Opus 4.8 (88.6%)  |  Gemini 3.1 Pro (~80%)

Top Open Source Models (SWE-bench Verified): DeepSeek V4-Pro (80.6%)  |  Llama 4 Maverick (~76%)  |  Qwen3-72B (~74%)

Top Small Models (15–50B): Gemma 4 31B (85.2% MMLU-Pro)  |  Qwen3-32B (~80%)  |  Phi-4 14B (~79%)

Top Edge Models (0–15B): DeepSeek V4-Flash (13B active)  |  Gemma 4 12B  |  Phi-4 Mini

AI Leaders (market cap): NVIDIA $4.7T  |  Alphabet $4.34T  |  Microsoft $2.83T

Robotics Leaders (market cap): Intuitive Surgical $151B  |  ABB Group ~$60B  |  Fanuc Corp ~$35B