The 'Need for Speed' in Mainstream AI Models — OpenAI 5.5 Instant, Google Gemini Flash, Anthropic Orbit. ARD-71
...+ Lidar's new uses beyond cars. AI Ramblings Daily #71 on AI-RTZ
The frame running through every item today: The ‘need for speed’ in mainstream AI models.
Beyond reasoning capability and AI Agent reliability, speed is becoming the next consumer wedge. AI Reasoning and AI Agents take time to compute the best reasoned steps and answers to user prompts.
Increasingly, the main model companies are trying to shorten that time of response — with friendly, on-point answers, delivered as ‘instantly’ as possible. Especially on the mainstream consumer front.
Today’s three items map the early moves by OpenAI, Google, and Anthropic to claim a position in that speed race. Plus a Gadget AI on Lidar (Light detecting and sensing) sensors finding cool uses way beyond self-driving cars.
Three Key Takes today:
(1) OpenAI’s GPT 5.5 Instant — The New ChatGPT Default Model. OpenAI is rolling out GPT 5.5 Instant as the new default model for ChatGPT — pitched as “smarter, clearer, more personalized.” Free for the free tier (likely with a few ads ahead), with the goal of pulling free users into paid subscription tiers ahead of OpenAI’s year-end trillion-dollar mega-IPO. The reporting: OpenAI — GPT 5.5 Instant. The TechCrunch angle: TechCrunch — OpenAI releases GPT 5.5 Instant, a new default model for ChatGPT. Standing thesis on OpenAI’s stormy weather lately: AI-RTZ #1071 — OpenAI’s Stormy Weather.
MP Take: “This is OpenAI’s fastest foot forward to increasing its 900 million weekly users of ChatGPT — a metric that has been stuck at that level for over three months. It’s caused a fair bit of consternation amongst investors and observers.
OpenAI is counting on this more ‘instant’ offering of its latest 5.5 model to provide fast utility to free tier users, giving them more reason to explore the paid subscription tiers that lead to more revenues ahead of its year-end trillion-dollar mega-AI IPO. The pitch for 5.5 Instant is ‘smarter, clearer, and more personalized.’ All for free, and likely a few ads.
This is leveraging its growing AI compute that they’re spending tens of billions of dollars on. Especially as OpenAI starts to deliver AI Agent type of functionality as well. All of this is to prep their metrics ahead of Anthropic — both are also in this mega AI IPO race targeted for the end of the year. And of course, both compete against Elon Musk’s SpaceX/xAI, which is shooting for its own IPO as early as June, around Elon’s birthday, at a valuation of $1.75 trillion and above.”
(2) Google Gemini 3.2 Flash Expected at Google I/O 2026 (5/19/26). Google is expected to debut Gemini 3.2 Flash at its annual developer I/O conference later this month — plus a fresh Gemma 4 open source release. The faster ‘flash’ Gemini fits squarely with Google rolling AI out via Gemini broadly to billions of users of all types of Google services — Search, Maps, YouTube, Docs, Drive, Gmail, and more. The reporting: Build Fast With AI — Gemini 3.2 Flash Release 2026. The Gemma 4 angle: Google — Developers Tools — Gemma 4 open source. Standing thesis on Gemini rolling out as the personal AI surface: AI-RTZ #1029 — Google Gemini AI Rolls Out Personal.
MP Take: “Google likely has a strong slate of AI upgrades and offerings planned for its annual developer I/O conference this month. And a faster ‘flash’ Gemini is part of the parcel. Especially as it has been rolling out AI via Gemini broadly to billions of users of all types of Google services, ranging from Search to Maps to YouTube, Docs, Drive, Gmail, and more.
Google is also leveraging its industry-leading TPU vertical stack infrastructure to serve up faster AI, as well as AI-optimized ads for its Gemini services. Each Google property has a billion-plus user base — Google has over half a dozen of these. The race here is to deliver useful utilitarian functionality in milliseconds or even faster speeds.
Google leverages deep integration possible with its services, unlike OpenAI counting on everyone coming to the ChatGPT app and typing their needs and requirements. Expect Google to press on these advantages, especially in partnership with Apple and Siri later this year.“
(3) Anthropic’s Orbit Proactive Assistant + SpaceX/xAI Compute Deal. Anthropic is debuting Orbit — a proactive AI assistant for Claude Cowork — at this week’s Code with Claude conference. Plus Anthropic announced an AI compute deal of convenience with SpaceX/xAI’s Colossus AI Data Center this week to absorb growth that’s outrunning its current capacity. The reporting: KuCoin — Anthropic’s New Proactive AI Assistant Orbit to Debut at Code with Claude Conference. The compute angle: Axios — Musk Anthropic Compute SpaceX AI. Standing thesis on Anthropic’s trillion-dollar momentum: AI-RTZ #1079 — Anthropic & OpenAI: Half of $2 Trillion in Cloud Co Backlogs.
MP Take: “Anthropic is on a tear with Claude Code and Cowork, with Cowork focused on mainstream users beyond coders and enterprise users. It’s racing to a $30 billion ARR run rate, even ahead of OpenAI a little bit of late.
It’s akin to Facebook leveraging its social network beyond .edu university users to mainstream users in the 2000s. And it’s a move that could make Anthropic a bigger consumer AI company ahead of OpenAI. Anthropic in many ways is a bit of a sleeper consumer AI company. Both directly and indirectly, Claude services are reaching consumers by the hundreds of millions. Perplexity uses Claude as a core model under its services. Increasingly, a lot of other companies are layering search and agent capabilities on top of Claude.
The company is seeing more growth than it can currently handle, leading it to do AI Compute deals of convenience like the SpaceX/xAI Colossus AI Data Center deal this week. Expect Orbit and other Claude Cowork features to deliver faster results of more utility to more mainstream users — especially against OpenAI, its arch nemesis as it were, and both of them against Google.”
Plus: Gadget AI — How Lidar Finds Uses Beyond Self-Driving Cars. Lidar — the laser-based depth-sensing tech that powers Waymo and other premium self-driving stacks — is finding uses way beyond autonomous vehicles. Shipping logistics, robotics, construction, surveying, agriculture, security. The technology that differentiates Waymo from Tesla on the autonomous-vehicle side is now expanding into broader physical-world AI applications. The reporting: NY Times — Autonomous Vehicles Technology Other Uses. Standing thesis on Waymo + Lidar: AI-RTZ #985 — Latest on Google Waymo’s Funding.
MP Take: “Lidar remains a key differentiator in self-driving cars vs Tesla in particular. Expanded uses are a plus for AI in the physical-world arena of opportunities. As AI takes over more applications in the physical world, these types of technologies are increasingly important — not just for sensors, but incorporated into things like robots to make sense of the world around them, gather more data, and generate more large language and small language model training.
What we need are less expensive sensors, and Lidar has come a long way from the big bulky things on top of cars a few years ago to now increasingly solid-state gizmos that can provide a lot more functionality. The Lidar advantage isn’t just safety theater — it’s a meaningfully different sensing modality from camera-only stacks, and that difference shows up in edge cases that camera systems struggle with.
As Lidar costs come down, expect the technology to spread into more physical-world AI applications — robotics, construction, agriculture, security — well beyond the autonomous-vehicle hero use case.”
Today’s AI-RTZ substack post #1079 covers Anthropic and OpenAI representing roughly half of the $2 trillion revenue backlog across Amazon, Microsoft, Google, and Oracle — the four biggest US cloud providers. The compute side of the AI mainframe stage is concentrating in a way the chip side already did, with interesting implications for both companies as they race toward their own mega AI IPOs later this year.
Closing Questions —
Most unexpected story on Lidar of late? Waymo founder Dmitri Dolgov’s story of a Lidar-equipped Waymo ‘seeing’ the feet of a pedestrian in front of a camera-blocking bus. He’s been on a number of podcasts and conferences telling this remarkable story. One of his Waymo cars in San Francisco used Lidar to detect a pedestrian on the other side of a bus — a person the cameras could not see because they were obviously blocked by the bus. But the Lidar sensor saw, as it were, the pedestrian’s feet under the bus. That signal provided some emergent capabilities to the Lidar AI, allowing it to wait for the pedestrian to come into view and cross safely before the car moved on. A remarkable unexpected application of these sensors that hadn’t been anticipated. It’s a vivid illustration of why Lidar’s not just a redundant sensor — it sees through occlusions that cameras can’t.
MP’s take on Lidar in self-driving cars per se? Additional sensors make ample sense vs the cost-saving imperative of Elon Musk’s Tesla. From MP’s perspective as a gadget freak for decades — we’ve been trained to buy MORE technical features than less. So this whole idea of buying a self-driving car without Lidar when prices for these things have come down to be par-or-close makes less sense. A Tesla here with just camera sensors (typically seven or eight of them) versus another car — Honda, Kia, any number of other cars that have Lidar sensors — at about the same price — and you’re getting an additional layer of sensors with additional capability that can see through, especially inclement weather. Cameras work great when visibility is great, especially during the day. In the evening, they’re less effective versus other sensors that are light-emitting, especially in inclement weather. From MP’s perspective, the notion of choosing technology with less tech features in it for the same price makes no sense.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)
Clips from today’s episode
Short — Waymo’s Lidar Saw a Pedestrian Through a Bus Waymo founder Dmitri Dolgov shared a remarkable story: one of his Waymo cars in San Francisco used Lidar to detect a pedestrian on the OTHER side of a bus — a person the cameras couldn’t see because they were blocked by the bus. But the Lidar sensor saw the pedestrian’s feet UNDER the bus.
MP Take: That signal provided emergent capabilities to the Lidar AI. The car waited for the pedestrian to come into view and cross safely before moving on. A remarkable unexpected application of these sensors that hadn’t been anticipated. It’s a vivid illustration of why Lidar isn’t just a redundant sensor — it sees through occlusions cameras miss.
Short — Google’s TPU Stack Is Why It Wins on AI Speed Google Gemini 3.2 Flash is expected to launch at I/O 2026 (5/19/26). Plus a Gemma 4 open source model.
MP Take: Gemini 3.2 Flash matters because Google leverages its unique vertical stack of TPU infrastructure to roll out Gemini-delivered AI services to billions of users. Both through the Gemini app, and Gemini AI embedded into a host of Google apps — Gmail, Docs, Drive, YouTube, Search, Maps. Each with a billion-plus user base. Google has over half a dozen of these. The race is to deliver useful utilitarian functionality in milliseconds or faster. Google’s deep integration with its services beats OpenAI counting on everyone coming to the ChatGPT app.
Short — Buying a Self-Driving Car Without Lidar Makes No Sense Tesla bet camera-only on self-driving years ago, mostly because Lidar sensors were expensive. The rest of the field — Honda, Kia, others — went multi-sensor with Lidar.
MP Take: As a gadget freak for decades, we’ve been trained to buy more tech features than less. The notion of buying a self-driving car without Lidar when prices have come down to parity makes no sense. Tesla with just cameras vs a Honda, Kia, or other car with Lidar at the same price — additional sensors add capability that can see through inclement weather. Cameras work great when visibility is great. In evening, fog, rain, light-emitting Lidar sensors are more effective. Choosing less tech for the same price makes no sense.
Short — Anthropic Is a Sleeper Consumer AI Company Anthropic is racing to a $30 billion ARR run rate, even ahead of OpenAI of late. Cloud Code is on the developer side. Cowork is the non-coder AI agent service on top of Claude.
MP Take: Anthropic in many ways is a bit of a sleeper consumer company. Both directly and indirectly, Claude services are reaching consumers by the hundreds of millions. Perplexity uses Claude as a core model under its services. Many companies are layering search and agent capabilities on top of Claude. Anthropic has Orbit expected to roll out broadly in coming days — they hinted at Code with Claude conference. Orbit and speed-focused features will differentiate Anthropic, especially against OpenAI.
That’s our AI Ramblings Daily #71 for today. Thanks for joining us. Stay tuned.
About AI Ramblings Daily (ARD), and AI-RTZ
Both are daily. Both are free. Both are about AI. But they’re different mediums carrying different messages.
AI-RTZ is the morning text — a deeper written take on one idea, published by at least 5 AM EST. Today: post #1079 — Anthropic & OpenAI now with a trillion-dollar backlog in AI infrastructure.
AI Ramblings Daily is the afternoon video + podcast — my ad hoc takes and perspective on the day’s AI issues & news flow, around 12 minutes today, with short 1-2 minute clips for quick topic views. Today: episode #71.
Subscribe to either or both on michaelparekh.substack.com. They run as separate Sections you can opt into or out of.
Links used in today’s show (already embedded inline above; listed here for reference)
Take 1 — OpenAI’s GPT 5.5 Instant:
Take 2 — Google Gemini 3.2 Flash + Gemma 4 (at I/O 2026, 5/19/26):
Take 3 — Anthropic’s Orbit + SpaceX/xAI compute deal:
Gadget AI — Lidar beyond self-driving:
Companion text:
Speed-tier proliferation is the labs admitting that latency is now a product axis, not just a quality axis. OpenAI 5.5 Instant, Gemini Flash, and Anthropic Orbit shipping in the same window says the labs are segmenting customers by latency tolerance the same way cloud has segmented by region. The interesting follow-up is whether enterprise pricing tracks this: a Flash-tier price for low-latency workloads and a Pro-tier price for deep reasoning gives the labs a way to expand revenue without expanding compute footprint as fast.