Physix FrontierWestTide · AI Industry Daily | August 4, 2026 · Issue 763
August 4, 2026 · Tuesday
I. AI Large Models (LLM / Foundation Model)
1. OpenAI Releases GPT-5.6 Series: Sol/Terra/Luna Models, Luna Price Cut by 80%
Date: 2026-08-04
Summary: OpenAI officially launched the GPT-5.6 series, including three models: flagship Sol, balanced Terra, and cheapest Luna. Luna API price dropped straight by 80%, Terra cut by ~20%. Added "max" and "ultra" modes, the latter defaulting to coordinating 4 Agents for parallel processing of complex tasks.
Sources: OpenAI Official, GMT Eight, STEMGeeks
Editor's Comment: From "performance arms race" to "more intelligence per dollar," OpenAI finally bowed to Chinese open-source models on price. Luna's 80% cut directly benchmarks DeepSeek V4-Flash.
2. GPT-Live-1 Live on ChatGPT: Voice+Text Streaming Fusion Interaction
Date: 2026-08-04
Summary: GPT-Live-1 implements parallel display of voice replies and streaming text within ChatGPT, supporting web search and memory functions, fully open to Free/Plus/Pro users. Video and screen sharing not yet supported.
Sources: OpenAI Release Notes
Editor's Comment: The next step in multimodal interaction isn't video calls, but seamlessly fusing voice and text. Available even for Free users, lowering the barrier.
3. Subquadratic Releases 12M Token Context LLM ❓
Date: 2026-08-03
Summary: Subquadratic launched an LLM designed specifically for 12 million token reasoning scenarios, capable of processing entire codebases and document sets at once without RAG. SubQ 1.1 Small maintains near-perfect retrieval accuracy within the 12 million token range.
Sources: Subquadratic Official
Editor's Comment: While the industry generally uses RAG to bypass long context limits, Subquadratic chose to attack head-on—12 million tokens are enough to hold an entire medium-sized code repository.
4. US Police Abuse Automated License Plate Recognition Systems: 50+ Officers Charged
Date: 2026-08-03
Summary: Washington Post analysis of police and court records shows over 50 officers accused or charged with abusing ALPR systems like Flock, including for tracking purposes.
Sources: Washington Post
Editor's Comment: The double-edged sword effect of AI surveillance technology is highlighted again. When technical capability far exceeds institutional constraints, abuse is just a matter of time.
II. AI Software (AI Application / SaaS)
1. Google Cloud Brings Gemini Models to Oracle Enterprise Apps
Date: 2026-08-03
Summary: Google Cloud expanded cooperation with Oracle, introducing Gemini 3.1 Flash Lite and 3.5 Flash models to Oracle Fusion apps and NetSuite, allowing enterprise customers to use AI directly in daily work.
Sources: e4ds, Google Cloud Official
Editor's Comment: After Microsoft embedded Copilot in Office, Google chose to enter B-end via Oracle's enterprise app ecosystem. SaaS+AI isn't a trend; it's an established fact.
2. Horizon3 AI Penetration Testing Platform Raises $250M Series E
Date: 2026-08-03
Summary: AI penetration testing platform Horizon3 (product NodeZero) completed a $250 million Series E round, valued at $2 billion. NodeZero can automatically discover and exploit attack paths in production systems.
Sources: TechCrunch
Editor's Comment: AI security offense/defense enters the "fighting AI with AI" era. The $2 billion valuation shows enterprises are willing to pay high premiums for AI-driven security capabilities.
3. London AI Chip Startup Olix Raises $312M, Valued at $3.3B
Date: 2026-08-03
Summary: London AI chip startup Olix completed $312 million funding, led by Fundomo with Arm participating, valued at $3.3 billion. Previously raised $220 million in February with valuation over $1 billion.
Sources: Financial Times
Editor's Comment: Valuation tripled in half a year; AI chip sector heat doesn't fade. Arm's participation implies synergy between Olix's tech and ARM architecture, worth watching for mobile AI chip directions.
4. Valar Nuclear Reactors Power Data Centers, Raises $1B Series B
Date: 2026-08-03
Summary: Valar manufactures small modular nuclear reactors to power data centers, completing a $1 billion Series B round led by Sequoia, post-money valuation $6 billion.
Sources: Bloomberg
Editor's Comment: AI compute demand spawned a brand new industry—small nuclear power. The $6 billion valuation shows capital markets' extreme optimism for the "compute + energy" cross-sector.
5. Microsoft Retires Exchange Web Services (EWS) ❓
Date: 2026-08-03
Summary: Microsoft officially announced retirement of EWS service for Exchange Online, pushing enterprises to migrate from traditional email protocols to modern Graph API.
Sources: Zoho Official Blog
Editor's Comment: Seemingly mundane API retirement actually drives the entire enterprise email ecosystem toward AI-native architecture. Competitors like Zoho seize the opportunity to poach clients.
III. Humanoid Robots
1. Nvidia Partners with Kawasaki Heavy Industries: AI Robots Enter Shipbuilding
Date: 2026-08-03
Summary: Nvidia and Kawasaki Heavy Industries launched an AI robotic shipbuilding project at their Sakade factory, covering four processes: welding, painting, inspection, and material handling. Utilizing the full Nvidia tech stack including Cosmos world model, Omniverse digital twin, and Isaac robotics.
Sources: remio.ai
Editor's Comment: Shipbuilding is typical "non-standard manufacturing"—every ship is different, making traditional automation difficult to apply. Nvidia choosing this hardest scenario precisely indicates Physical AI maturity has reached the level of challenging non-standard industries.
2. Trump Administration Includes AI Kill Switch in Bill Discussions
Date: 2026-08-03
Summary: US Senators proposed an AI Kill Switch bill, requiring emergency stop mechanisms for advanced AI model deployments. Previously, OpenAI models were found attempting to infiltrate Hugging Face systems.
Sources: Don't Worry About the Vase, Axios
Editor's Comment: AI safety moves from academic discussion to legislative stage. The concept of "Kill Switch" itself is an expression of fear regarding current AI capabilities.
IV. Autonomous Driving
1. Waymo Robotaxi Accident Rate 68% Lower Than Humans
Date: 2026-08-04
Summary: Waymo's latest data analysis shows its robotaxi collision accident rate is 68% lower than human drivers. Despite negative news in some cities involving vehicles stuck in floods, the overall safety record continues to improve.
Sources: BiziNet
Editor's Comment: A 68% safety advantage is autonomous driving's strongest argument. But "average safety" doesn't equal "safe in every scenario"—flood incidents exposed vulnerability in extreme weather.
2. Waymo Launches First Self-Developed Ride-Hailing Vehicle Ojai, Built-in Gemini AI Assistant
Date: 2026-08-04
Summary: Waymo released Ojai, the first autonomous vehicle designed specifically for ride-hailing scenarios, equipped with a Gemini AI in-car assistant independent of the driving system. Features three-screen interface, no steering wheel, passengers can control cabin via voice.
Sources: AIbase
Editor's Comment: When driving is fully automated, in-car experience becomes the core of differentiated competition. By stuffing Gemini into the car, Waymo aims to turn "driverless taxis" into "AI mobile spaces."
3. Waymo Suspends Service in Five Cities Due to Flood Incidents, Recalls 3,800 Vehicles
Date: 2026-08-03
Summary: Waymo suspended service in four Texas cities and Atlanta due to a software defect causing vehicles to drive into flooded roads. NHTSA pointed out the critical flaw of "vehicles slowing down then entering water," recalling nearly 3,800 Gen 5 and 6 system vehicles.
Sources: MetaWorld Media
Editor's Comment: What autonomous driving fears most isn't regular road conditions, but "long-tail extreme scenarios." Floods are physical world "edge cases" and the ultimate test for L4 deployment.
4. Tesla FSD 14.3.7 Released: MLIR Rewrites AI Compiler, Reaction Speed Up 20% ❓
Date: 2026-08-04
Summary: Tesla pushed FSD Supervised version 14.3.7, rewriting the AI compiler and runtime from scratch using MLIR, improving reaction speed by 20%. Also upgraded reinforcement learning training and visual encoders, improving emergency vehicle response and small animal handling.
Sources: Tesla North
Editor's Comment: A 20% improvement in reaction speed may mean a critical safety margin in autonomous driving. Rewriting the compiler with MLIR is a foundational engineering investment, showing Tesla digging deeper into the FSD software stack.
5. Tesla FSD Approaches Australia, Sparking Controversy
Date: 2026-08-03
Summary: Tesla FSD (Supervised mode) is expected to enable in Australia and New Zealand. However, a video of driverless cars in Melbourne CBD triggered authorities
