SpaceX Completes Three Falcon Launches in Under 13 Hours as Starship Deploys 26 Starlink V3 Satellites

Key Takeaways

1. SpaceX completed three Falcon launches and four booster landings in under 13 hours, plus a Dragon docking for its 21st crewed mission.
2. Starship's 14th flight reached orbit and deployed 26 Starlink V3 satellites, with Musk calling the hot-staging sequence "science fiction made real."
3. Tesla reported Q3 2026 production of 464,391 vehicles, 486,532 deliveries, and 13.7 GWh of energy storage deployment.
4. Perplexity released a batch of open-source contributions including a multimodal decision model, a context embedding model, and a local inference engine for Apple silicon.
5. Supabase raised $150M at a $10.65B valuation, with over 13 million developers and 70% of new databases created by agents or AI-driven tools.
6. Vercel confirmed a KVM 0day via its Sandbox bug bounty program, described as a full VM escape in an industry-standard hypervisor.
7. Musk said roughly 90% of SpaceX revenue this year will come from commercial business, with government revenue under 5% in Q4.

1. SpaceX Launch Cadence and Starship Progress

1. SpaceX completed three Falcon launches and four first-stage landings across four landing zones in Florida and California in under 13 hours, while also docking a Dragon spacecraft to the ISS for the company's 21st crewed mission. A separate summary noted five missions in six days: USSF-385 on September 26, Starship's 14th flight on September 28, and Crew-13, Transporter-18, and NROL-97 on October 1, with three of those occurring within 13 hours. — via 1 2 3 4
2. Starship's 14th flight completed its first orbital flight and deployed 26 Starlink V3 satellites. Musk described the hot-staging scenario as "science fiction made real." A retweet projected each launch could generate over $700M in annual incremental revenue, growing as SpaceX AI enters AI software monetization. — via 1 2 3 4
3. Musk said roughly 90% of SpaceX revenue this year will come from commercial business, with government revenue under 5% in Q4. He contrasted this with federal spending at about 25% of the economy, arguing SpaceX is "far more commercial" than the average US company. He also cited a claim that SpaceX has saved the US government and taxpayers about $14 billion, saying the actual figure is now higher. — via 1 2

2. Tesla Operations and Products

1. Tesla reported Q3 2026 production of 464,391 vehicles, 486,532 deliveries, and 13.7 GWh of energy storage deployment. The company said it will update on October 21 at 16:30 CT via a livestream on X. — via 1
2. Musk said Robotaxi operations in Austin were extended from 10 PM to 11 PM, with the main issue being difficulty seeing pets at night to avoid running over gray kittens. Tesla's Supercharger network now has over 85,000 charging stalls across more than 9,000 stations. — via 1 2
3. NASA certified the Cybertruck as an astronaut emergency evacuation vehicle, allowing SpaceX to replace aging MRAPs. A NASA spokesperson said both parties evaluated vehicle capabilities across emergency scenarios and determined the Cybertruck meets NASA requirements. — via 1

3. AI Products, Open Source, and Infrastructure

1. Perplexity released multiple open-source contributions: pplx-decider-v1-27b, a multimodal decision model averaging 85.7% across 11 benchmarks and leading Jev; pplx-embed-v2-context-9b-preview, a context embedding model that performed best on ConTEB and turbopuffer context-bench; and Lily, a local inference engine for Apple silicon built in Rust with custom Metal kernels that does not depend on PyTorch or MLX, running 1.23x faster prefill and 1.35x faster decode than MLX-LM on M5 Max. Additional releases include PII-Tracer, a 0.6B on-device PII classifier that outperformed OpenAI's Privacy Filter on all five public benchmarks alongside a 13-language, 13,000-conversation PII-TRACE benchmark; WANDR, a breadth-and-depth research agent benchmark with 500 tasks requiring 170,000 sourced records; Numbat, an agent detection and response tool with 52 rules shipping as a single Go binary for macOS, Linux, and Windows; and Bumblebee, a read-only supply chain scanner for developer machines covering packages, MCP configurations, and editor and browser extensions. The author said more open-source contributions are coming. — via 1
2. Aravind Srinivas said the goal is to vertically integrate agent infrastructure by building its own sandbox and optimizing it for the best chips, arguing Vera is far superior to x86 and expecting to share more as Perplexity Computer begins deploying on Vera. He also retweeted anticipation for a collaboration with NVIDIA's Vera CPU on SPACE. — via 1 2
3. Supabase raised $150 million at a $10.65 billion valuation. More than 13 million developers use it as an application backend, with over 1 million new users and over 4 million new databases per month, and 70% of new databases created by agents or AI-driven tools. A separate retweet noted 60% of the latest YC batch uses Supabase. — via 1 2
4. Vercel confirmed a KVM 0day through its Sandbox bug bounty program, affecting what it called the industry gold standard for Linux virtualization. Guillermo Rauch said 2026 is wild and thanked researchers including Paulos for helping build the safest agent sandbox, with a full report coming soon. A quoted post described it as a full VM escape 0day from guest to host root in an industry-standard hypervisor. — via 1
5. Sam Altman said dot is his favorite OpenAI product, noting it visibly improves daily because it continuously learns his workflow and style, and that delegating tasks he dislikes—which normally accumulate into a "gravity well of dread"—has been a pleasure. He also responded to speculation about a Cerebras partnership, calling Cerebras a close partner with deep collaboration pushing the frontier of speed. — via 1 2
6. Sam Altman said he is very disturbed by attempts to give AI models religious-style power or to ask people to abandon human judgment, calling it a real safety concern. — via 1

4. Hardware, Manufacturing, and Developer Tools

1. Greg Isenberg outlined a "vibe manufacturing" workflow: measure precisely with digital calipers and phone LiDAR apps like Polycam, since inaccurate measurement is the top cause of first manufacturing failures; then have frontier models write code-based CAD in OpenSCAD or CadQuery, or use Zoo's Text to CAD for one-step generation. He stressed factories need exact file formats—STEP for machining, DXF for laser-cut metal, STL for 3D printing—and that text-to-3D models on Hugging Face look correct in shape but are not usable. He recommended having AI validate designs against actual factory processes such as metal bend tightness, minimum hole sizes, and part clearances, prototyping in plastic with a $300–$600 Bambu Lab printer, then paying a mechanical engineer on Upwork about an hour to review drawings as the cheapest insurance. — via 1
2. For production and electronics, Isenberg listed SendCutSend for cut and bent metal, Xometry and Protolabs for machining and plastics, and JLCPCB for circuit boards, typically shipping in days. Smart features often use cheap WiFi chips like ESP32, with boards designed in KiCad and controlled via ESPHome on a phone. He advised ordering one test unit before adjusting, factoring parts, finishing, assembly, packaging, and shipping into cost, and leaving enough margin since custom products are hard to resell. He recommended made-to-order with customer prepayment for zero inventory, safety certification for products that plug into walls or involve children, and said the slowness and hassle are the moat, with batch production cutting costs quickly once orders stabilize. — via 1
3. Nikita Bier said the era of customizing products to consumer preferences has arrived, citing his own "vibe-fabricated" dog door with a WiFi-controlled lock built to his home's specifications. He said he does not understand metalworking or electrical engineering, yet the product will be manufactured and delivered within two weeks at a cost nearly equal to buying an off-the-shelf mass-produced product. — via 1 2
4. Muse Gadgets launched as open-source ESP32 firmware and a Linux SDK for building hardware devices that work with Muse. Developers can get an API token from a specified address and point their usual coding agents at the GitHub repository to build Muse peripherals. Ryan Hoover commented that he loved it and it made him want to build something physical. — via 1 2 3
5. Garry Tan said Capy's cross-session coordination exceeded expectations: fix work initiated by its GBrain collaborator Sina avoids items he is handling across multiple Capy threads, and it is model-agnostic. He also said he deleted about a thousand lines of markdown from gstack because frontier models are now good enough that old tricks are generally unnecessary. — via 1 2
6. RogoAI engineers built internal company applications on Vercel, deploying over 73,000 times last month. Coding agents deliver to production within 5 minutes, covering use cases from churn risk analysis to a sales team's deal desk. — via 1

5. Policy, Society, and Notable Claims

1. Jack Mallers argued governments give people two choices: default on their debt or repay it with worthless paper money, and said it is never too late to buy Bitcoin. He also retweeted a claim that "your cash has a new income source," with a "soon" teaser. — via 1 2
2. The Indian government has formally removed bitchat from the App Store. A retweet from the same account stated that no one can inflate Bitcoin. — via 1 2
3. Amjad Masad commented "Reaching concerning levels of psychosis" on a report that, according to The New York Times, Anthropic co-founder Chris Olah threatened to withdraw from a May AI encyclical launch event because Pope Leo XIV refused to acknowledge machines might have consciousness. The report said his team then privately lobbied the pope's advisers to "take seriously the possibility of model consciousness," the pope did not concede, and Anthropic had for months hosted theologians and religious scholars under NDAs hoping they would recognize Claude as having moral status. This item is disputed/unverified. — via 1
4. A Stanford professor said AI caused students to get all A's on homework and never attend office hours, yet fail in-person final exams. The department canceled office hours and instead required TAs to meet one-on-one with students to walk through each programming assignment and prove they truly understand the code. — via 1
5. Brian Chesky endorsed a post arguing Spain's housing problem is not a lack of decrees but a lack of housing: subsidizing demand without increasing supply raises prices, over-regulating a market without new construction manages scarcity rather than eliminating it, and taxing a rental form changes incentives without adding a single brick. The post called for more usable land, permitting cut to months instead of years, higher density where possible, legal certainty, and a large increase in public and private housing. — via 1
6. Paul Graham retweeted a claim that a Stanford professor's experience shows AI is masking student comprehension, and separately retweeted that Starship's first orbital flight delivered 26 Starlink V3 satellites. — via 1 2