Big Battery Health update, live now! ⚡🔋 • Way faster across the board • Squashed a bunch of bugs • Sharing your numbers is so much cleaner • Smoother flow start to finish The honest number on your Tesla's battery how you stack up against the whole fleet. Give it a spin! 🔥 iOS now, Android soon 👉 batteryhealth.app Please consider giving us a 5 star review - apps live and die with 5 stars. Thank you. @teslaownersSV @DirtyTesLa @tesla_raj @Tesla

The metric I keep coming back to for SpaceX is $/Mbps to orbit Starlink exists because Falcon 9 dropped bandwidth deployment costs ~10x to ~$6.55/Mbps. That’s about to drop again to just $0.30/Mbps because of Starship. A business that is doubling users annually with a 63% adjusted EBITDA margin is about to cut their biggest cost by 95%… It really seems like people don't understand the implications of this. The math assumes a reusable Falcon 9 launch is 17 tonnes at $1,000/kg and 2,600 Gbps per launch. Starship is targeting 100 tonnes at under $185/kg and 61,000 Gbps per launch. That's $17M for 2,600 Gbps ($6.55/Mbps) verse $18.5M for 61,000 Gbps ($0.30/Mbps). Starship's additional volume allows for larger satellites, enabling simultaneous gains on multiple cost curves. The math suggests V3 satellites are ~600 Mbps/kg vs ~150 Mbps/kg from V2 mini. Combining the 4x improvement on satellite bandwidth density with a 5x improvement in launch gets you the 20x improvement to 30 cents per Mbps to orbit. These are fairly conservative assumptions because launch probably comes in even lower as Starship ramps, and satellite improvements probably keep coming. At $0.10 / Mbps, $1 billion spend on launch represents 10,000 Tbps or about 15x the bandwidth of Starlink's constellation today. $1B is 90 days of operating income for Starlink... at it's current scale... Yeah, I really don't think people are getting this. Starlink is the internet now.

Hope we can help make it happen.

Day three of amplifying this. If this interview doesn't happen I am legitimately going to crash out.

Everyone's arguing about FSD with vibes and screenshots. We built the receipts. 🚗🔥 TezLab now runs the most detailed FSD analysis anywhere — our own proprietary system that breaks down every drive, mile by mile. Not just "how far did FSD drive me" — but how much YOU stepped in. 👇 — Every drive, you get: 📍 FSD distance % of the drive ✋ Go-pedal overrides (time distance) 🛑 Braking & interventions, segment by segment 🗺️ The whole route mapped with an event timeline Watching your own numbers improve over time is honestly addicting. 📈 — FSD analysis lives in TezLab Pro — start a free trial, grab some FSD credits, and run your drives through it. iOS & Android. 👉 tezlabapp.com/subscribe @wholemars @DirtyTesLa @teslaownersSV @elonmusk @robotaxi @TesLatino

Amazing event. Get there somehow.

📷Memorial Day Sale $10 off your first year of TezLab Pro📷 Save $10 on Your First Year of Pro To mark the long weekend, we're taking $10 off your first year of TezLab Pro. Upgrade by May 31 to unlock unlimited drive and charge history and every Pro feature. Available only at tezlabapp.com. Get your first year of TezLab Pro for $10 off. The discount is applied at checkout on our website — no promo code needed. tezlabapp.com @DirtyTesLa @wholemars @TesLatino @teslaownersSV

Tesla Users… who use @TezLabApp? I do. Would love to create a 𝕏 Tesla Fam group of all of you in it. Who would be interested? Here’s the group link… tezlab.group/tesla-fam @Tesllatino @wholemars @DirtyTesLa i know you do. I love this app.

My team at @Tesla is hiring a talented 3D generalist to help build the Tesla App! Looking for a strong portfolio of 3d art, experience with game engines, and an eye for design You’ll get to ship to real customers, build cool stuff, and work closely with me

We updated the entire @TezLabApp backend. 🔥 Months of heads-down work that doesn't show up in a release notes page — a complete update. As of this week it's all live, and you should feel it the moment you open the app. If you have a lot of TezLab history (and so many of you do), you know what I'm talking about — the spinner, the slow charger map, the FSD breakdowns that took a beat. That's mostly gone now. 🚗 Trip data loads instantly 🤖 FSD analysis crunches in seconds 📍 Road trip planning is alive ⚡ The charger map populates the moment you pan 🔥 The Zaps feed actually keeps up 🚀 And every smaller thing too Plus — we just launched a brand new app: Battery Health for Tesla 🔋 The honest number for your battery, max range plotted over time, and how you stack up against every similar Tesla in the TezLab fleet. iOS now · Android coming soon → batteryhealth.app Already a TezLab Pro customer? You already have all of this in the main app. Free 14-day trial: tezlabapp.com/subscribe Founders Series (lifetime access): tezlabapp.com/founders Cheers — Ben and the larger TezLab team. @DirtyTesLa @wholemars @TesLatino

We've been hard at work building what we do best, the absolute best in class apps. This time, Battery Health. There's always so much speculation around battery health, so we wanted to wrap it up in a nice, dedicated experience. This is also in the core TezLab app, but if you're new to the community or want a beautiful, focused app, look no further than Battery Health For Tesla. Share the love and spread some sugar (reviews on it). Simple and beautiful - work of art. And if anyone was actually in doubt, Tesla Batteries Are Incredibly Durable! @wholemars @DirtyTesLa @elonmusk @teslaownersSV @ThomasAlxDmy batteryhealth.app/

We’ve learned how to build skilled agents, but still don’t know how to run them as a team. Huawei Noah’s Ark Lab proposed OneManCompany (OMC) – an organisational layer for multi-agent systems that treats agents like employees in a real company. The main idea is to move from Skills → Talents A Skill is a tool/function inside one agent. But a Talent is a full agent package: role, tools, skills, prompts, configs, and working principles. So OMC can "hire" a specialist agent. The system has 3 main pillars: 1. Talent Container = Employee Talent = who the agent is: developer, designer, researcher, etc. Container = where it runs, like Claude Code, LangGraph, scripts, etc. So different kinds of agents can work together in one organisation through shared interfaces. 2. Talent Market If the current team lacks a capability, OMC can recruit a new "specialist" from a marketplace of verified agents. The agent team is fully configurable 3. E2R tree search: Explore–Execute–Review loop - explores how to decompose the task - assigns subtasks to agents - executes them - reviews outputs before they can unblock downstream work - retries or re-decomposes when something fails Tasks form a dynamic tree/DAG with review gates, so bad intermediate outputs don’t silently propagate. Plus, agents update their own working principles after tasks and feedback. There is even an HR-style lifecycle: performance reviews → Performance Improvement Plan (PIP) → offboarding → replacement from the Talent Market. OMC proved its efficiency, reaching 84.67% success on PRDBench, about 15 percentage points better than previous baselines. So this is a strategy worth taking a closer look at.

Great read.

FSD 14.3 is two weeks old in our fleet data. Too early for conclusions, but the behavioral signal is clean. Go-pedal override (median): 1.8%. That's the lowest of any 14.x sub-version we've tracked. For context, 14.2.2.5 sits at 2.8%, 14.2.2.4 at 3.6%. Hard braking: 10.8 per 1,000 miles vs 15-18 for mature 14.2.x builds. Whole-drive autonomy is right at 50%, in line with the broader 14.x average. The caveat: small sample, early adopter bias is real, and two weeks of data isn't a trend. But when go-pedal AND hard braking AND whole-drive all point the same direction from day one, it's worth paying attention. We'll keep tracking. Gigalytics data. @TezLabApp @munster_gene @wholemars

The FSD subscription flywheel is real, and nobody's tracking it. Here's what's actually happening inside Tesla's FSD business right now, from our fleet telemetry data at @TezLabApp. The metric nobody's watching: subscribed % of fleet. Oct '25: 2.42% Jan '26: 3.56% Apr '26: 5.24% That's more than doubled in six months. And it's accelerating. Subscribed share added almost a full point in March alone (4.34% → 5.13%). To understand why, you have to go back to what Tesla did over the holidays. In late November 2025, Tesla pushed a free 40-day FSD v14 trial to roughly 1.5 million HW4 vehicles. Biggest trial deployment they've ever run, timed perfectly for holiday road trips. In our data, December's new subscriber cohort was 84% trial users, just 16% paid. Tesla was betting that seat time would convert free users into paying customers. That trial expired January 8. What happened next is the interesting part. Trial volume collapsed. December's cohort was overwhelmingly trial, but January flipped to 78% paid. By March: 97% paid. April is tracking 96%. The trial-as-funnel strategy didn't produce mass conversion. But what it did do is clarify the subscriber base. Look at M1 retention (the percentage still subscribed after one month) by subscriber type: Paid M1 retention is consistently in the 86–100% range across cohorts with meaningful sample sizes. Remarkably stable. Trial M1 retention: 81% in October, 41% by March. And by M2 (two months), trial retention drops to single digits. The December holiday cohort, Tesla's biggest bet, retained just 1% at M2. Essentially nobody who got the free trial stuck around. Here's the thing people will miss. The overall M1 retention number for March looks great at 91.1%. But it looks great because the composition changed, not because retention "improved" in the abstract. When 97% of your cohort is paid subscribers who retain at 92% , the blended number naturally rises. That's not a retention improvement. It's a base quality improvement. Important distinction. Then on February 14, Tesla ended outright FSD purchases entirely. No more $8,000 one-time buy option. It's $99/month or nothing. (as of today:) This is where the compounding kicks in. The mechanics: new paid subscribers enter each month. They retain at 87% even four months in (the October '25 paid cohort, our oldest with M4 data, is still at 87%). Very few leave. Each new month's cohort stacks on top of the ones still there. Since the trial churn cleared in late January, activations have exceeded churns every single week. 12 straight and counting. That's how you get subscribed share doubling. Meanwhile purchased % peaked around 8.5% in December and has been slowly eroding. 7.78% and declining. Natural attrition as people sell vehicles, trade in, etc. Since Tesla killed the purchase option, that line can only go down from here. The subscribed line can only go up. At the current trajectory, they cross sometime this year. (A few older subscribers transferred FSD to HW4 but that's a very small number.) One more corroborating signal worth paying attention to: the people who subscribe aren't casual about it. FSD now accounts for 50.5% of driving distance and 63.1% of driving duration among users. These aren't people toggling it on occasionally. FSD is their default driving mode. The gap between distance and duration is itself interesting. FSD overindexes on city and slower-speed driving versus highway, which is where the higher duration percentage comes from. That's consistent with what we see in the whole-drive autonomy data: v14.x users complete 50% of their drives entirely on FSD without intervention. What this doesn't tell you: whether the holiday trial "failed." You could frame 1.5 million free trials with near-zero conversion as a waste. Or you could frame it as Tesla stress-testing the funnel at massive scale and learning that the product now sells itself to people who are ready to pay. No free taste needed. The fact that Tesla followed the trial by killing outright purchases suggests they're betting on the second interpretation. Either way, the observable outcome is this: the casual users washed out, the purchase option is gone, and what's left is a subscription base that compounds because retention is high enough that inflows consistently beat outflows. That's the flywheel. And it's the number to watch going forward. Exclusive data from @TezLabApp fleet telemetry · gigalytics.io @CernBasher @herbertong @wholemars @DirtyTesLa @farzyness

Neural Computers explained It looks like a computer running code, but it’s just a model rendering the simulation. No real computation What is actually happening here - and is this the future of computing or just a very convincing illusion?

Getting so good.

Tesla Q1 Inventories - Relax People By my calculations, Tesla's global vehicle inventory was ~28 days for Q1, up from just 15 days in Q4. But 28 days of inventory is not the true picture - it's really lower than that if we take into account "transport days." 1) Transport Days: the number of days needed to transport newly produced vehicles from the factory to the end customer. If we assume that vehicles made in Fremont, Texas and Berlin generally take two weeks to be delivered via rail and/or truck to the customer, and if we assume that vehicles made in Shanghai are split 50/50 between exports and in-country sales then the global average transport time comes to about three weeks (measured as 18 days). 2) True Inventory Days: this is the difference between the total Inventory Days of Supply and the Transport Days - bringing the Q1 2026 number to just 10 days of inventory! This is a global average. In reality the inventory is likely even lower in the US and higher internationally. @thejefflutz