XAI Opens Grok 4.5 to the Public as SpaceX, Tesla, and X Stack Fresh Operating Data

A coding-first model at $2 and $6 per million tokens lands against OpenAI and Anthropic, while a $50 million Starship lunar cargo booking, a Model Y China sales crown, FSD incident rates, a 2040 net-zero chain target, and an X-Money Visa card all hit the same news cycle.

On July 8 and 9, 2026, the Musk stack put product, logistics, demand, safety, and payments on the table in one burst. xAI pushed Grok 4.5 into public use as a coding and agent model trained with Cursor, priced and speed-tuned for real developer work. SpaceX showed up as lunar logistics infrastructure through iSpace’s $50 million Starship cargo booking. Tesla put hard China registrations, FSD incident rates, and a full-chain 2040 emissions target in front of the market. X moved payments from abstract plan to a metal Visa card in selected hands. The through-line is operating proof, not slogans.

Key Takeaways

• Grok 4.5 went public on July 8, 2026 as xAI’s first model trained specifically for coding, agent tasks, and knowledge work with Cursor, served at 80 tokens per second.
• Token pricing sits at $2 per million input and $6 per million output, with 4.2 times fewer output tokens than Opus 4.8 Max on SWE Bench Pro tasks.
• iSpace booked 500 kilograms of Starship lunar payload capacity valued at $50 million for a landing no earlier than 2030.
• Tesla’s Model Y led all passenger vehicles in China in June 2026 with 38,654 retail registrations, including over every fuel type, while sitting above every other top-10 price.
• Brand-level China retail was 52,920 units, 3.3% share, and a 14% year-over-year decline even as Model Y outsold the runner-up by 5,295 units.
• FSD Supervised logged 0.19 major incidents per million miles versus a U.S. average of 1.51, with minor and off-highway collisions reported 7x and 6x lower, on a fleet of roughly 9 million vehicles and more than 11 billion miles.
• Tesla is targeting net-zero emissions across the full value chain by 2040, with 100% renewable electricity well before that date, after Berlin’s third straight year on fully renewable power.
• Sawyer Merritt received a physical metal X-Money Visa debit card; beta terms reported elsewhere include partner-bank deposits, 6% APY on balances, and 3% cash back on eligible purchases.

---

Grok 4.5 Enters the Coding Agent Fight

xAI released Grok 4.5 for public use on July 8, 2026 as its first model built around coding, agent work, and knowledge tasks with Cursor in the training mix. The company is pitching speed and cost as product features: 80 tokens per second, 4.2 times fewer output tokens than Opus 4.8 Max on SWE Bench Pro, and list prices of $2 per million input tokens and $6 per million output tokens. That package is aimed straight at OpenAI, Anthropic, Google, and specialized developer tools, not at abstract chat benchmarks alone.

The infrastructure story sits next to the model card. A custom C and C++ inference stack mapped to new NVIDIA Grace Blackwell hardware is framed as a path that could roughly double current speed, with the build harness improving almost daily. Beta feedback was treated as strong enough to leave private testing. European Union availability is still expected in mid-July, so one major region is not fully open yet.

The numbers do not say the model already owns every leaderboard. Company materials still show Grok 4.5 trailing Fable and some GPT 5.5 variants on parts of the Egenic and software-engineering set. Pareto dominance talk on coding utility only matters if developers keep paying for it inside real workflows. The next checkpoints are paid adoption, measurable latency gains, the EU flip, and a 2-trillion-parameter model Musk has said finishes training this month. I think the bet is tight coupling of model quality, IDE workflow, and custom inference. If latency and cost keep improving week by week, that compounds inside agent loops. If not, price and speed become marketing, not a moat.

Starship as Lunar Logistics Capacity

iSpace said it will run a lower-cost lunar cargo business on SpaceX’s Starship rocket and moon lander, with 500 kilograms of payload capacity on a mission targeted no earlier than 2030. The agreement is valued at $50 million. That is rideshare language for the Moon: third parties package and sell capacity rather than every customer flying a dedicated stack.

iSpace is building a mobile cargo system to integrate, move, and support smaller customer payloads on the surface, positioning itself as a lunar access integrator for global customers with relatively small delivery needs. High capacity at relatively low cost is the stated condition for a sustainable lunar economy. The commercial signal is real even when the kilogram count is modest. Companies are starting to book lunar logistics the way they book launch.

Execution risk is still large. Starship has not demonstrated commercial lunar cargo landings at scale. The schedule runs into the next decade. iSpace is coming off difficult early lander missions. Shared delivery only works if demand shows up and the surface vehicle ships on time and budget. I would watch a confirmed mission manifest, customer pricing, the mobile cargo timeline, and any SpaceX language on commercial lunar cargo availability. A single 500-kilogram booking does not make a market. It does show Starship priced as infrastructure, not only as a rocket demo.

Model Y Takes China’s June Title at a Premium

Tesla’s Model Y was China’s best-selling passenger vehicle of any fuel type in June 2026, with 38,654 retail registrations. It ranked first while priced above every other vehicle in the top 10. China is Tesla’s most competitive EV market, and domestic brands have kept pricing pressure high, so an all-powertrain monthly win is a concrete demand signal, not a vanity ranking.

Broader brand data softens the victory lap. Total Tesla China retail sales were 52,920 in June, good for 3.3% market share and a 14% year-over-year decline. Model Y outsold the second-ranked model by 5,295 units, yet May had Model Y second overall behind Geely Xing Yuan. One strong model does not automatically fix portfolio pressure.

I think this complicates the story that Tesla is simply losing competitiveness in China across the board. The result only sticks if it holds over several months without deeper incentives that crush margin. Technology leadership, pricing power, and margin still matter more than unit count in every segment. Watch whether official CPCA figures line up with retail data, and whether volume holds without another round of price war.

FSD Supervised Incident Rates Versus National Averages

Tesla vehicles using FSD Supervised logged 0.19 major incidents per million miles against a U.S. average of 1.51. Minor collisions were reported seven times lower and off-highway collisions six times lower than the comparison set. The company’s safety materials point to roughly 9 million vehicles and more than 11 billion miles of driving behind the broader fleet context. FSD Supervised still requires an attentive driver and does not make the car autonomous.

Those rates sit where regulators, insurers, and buyers make trust decisions. Lower incident claims can support autonomy features as a product advantage long before robotaxi scale, and they can feed insurance cost and paid-ride trust arguments. Directionally, the gap versus national averages is large enough that the market will keep using it.

Methodology still decides how much weight the numbers carry. Incident definitions, exposure mix, weather, city versus highway miles, vehicle type, and driver behavior all change the comparison. I still want third-party validation. Safety data matters most because regulation is the main scaling bottleneck for autonomy. The next useful signal is finer methodology disclosure and whether insurers or regulators put the data into filings, approvals, or pricing.

Full Value Chain Net Zero by 2040

Tesla is targeting net-zero emissions across the full value chain by 2040, covering suppliers, manufacturing, product use, and end-of-life, with a plan to move operations to 100% renewable electricity well before that date. Gigafactory Berlin-Brandenburg has run on fully renewable electricity for a third straight year. Steel and battery materials remain the hard-to-abate categories in the company’s own framing.

A chain-wide target pulls suppliers, factory capital, battery sourcing, and energy procurement into one operating constraint. It also tightens the link between the car business and Tesla Energy as data-center power demand rises. The ambition is real. Steel and batteries do not decarbonize on a press release.

Initial reporting does not spell out detailed interim milestones, supplier contracts, or third-party verification. Investors have to separate a long-dated commitment from measurable steps. The useful watch list is supplier requirements, Scope 3 reduction pathways, and progress on steel and battery materials. I treat 2040 as a capital-allocation and sourcing constraint, not as a brand slogan.

X-Money Leaves the App and Hits a Wallet

A physical metal X-Money Visa debit card is now in at least one public beta user’s hands. Business coverage has framed X-Money as a debit card and digital payments platform built with Visa. Reported beta terms include deposits through partner banks, 6% APY on balances, and 3% cash back on eligible purchases. That package is designed to pull the product into everyday spending, not only in-app transfers.

Payments sit at the center of X’s push past advertising and subscriptions. A metal card is a distribution and habit test more than a hardware story. Financial terms can seed early users. Habit, trust, and unit economics decide whether the product survives contact with daily spend.

X has not published a full public launch schedule, adoption metrics, partner-bank terms, or the full regulatory map. The business case still depends on availability, compliance, fee economics, and repeat use beyond selected U.S. Premium Plus users. Watch a public launch date, state-by-state availability, bank disclosures, and evidence of usage that is not just a novelty photo of the card.

One News Day, Shared Operating Theme

The same 24-hour window stacked a public coding model, a lunar cargo booking, a China sales ranking, FSD incident rates, a 2040 supply-chain climate target, and a payments card. That is a lot of surface area. The shared theme is operating data and commercial packaging: tokens and latency, kilograms and dollars of lunar capacity, retail registrations, incidents per million miles, renewable factory power, and yield plus cash back on a debit product.

I would keep the scoreboard simple. For xAI: developer adoption and week-by-week inference gains. For Starship: manifest, pricing, and cadence that can support routine lunar logistics. For Tesla: multi-month China durability without margin-killing incentives, methodology depth on safety, and interim Scope 3 execution. For X: real launch, partner banks, and daily use. Busy days are cheap. Follow-through is the product.