GT Axis Daily, July 13, 2026

GPT-5.6 comes in three tiers: Sol for heavy lifting like coding and research, Terra for everyday work at roughly half the cost, and Luna for fast, cheap, high-volume tasks. Think of it as choosing between a sports car, a sedan, and a scooter from the same maker, different jobs, different price points. It's rolling out now across ChatGPT, OpenAI's coding tool Codex, and its developer API, with full availability in the coming weeks.

Musk called it "Opus-class," meaning it goes toe-to-toe with Anthropic's top-tier model, but says it's faster and cheaper to run per task. It was trained in part using data from Cursor, the coding tool SpaceX is now buying outright (more below), a preview of how tightly these two companies are about to merge.

The new unit, Microsoft Frontier Company, embeds 6,000 engineers and industry specialists directly inside client companies to help them actually get value out of AI instead of just buying the software and hoping. It's a quiet admission that most companies are struggling to turn AI spending into real returns, and Microsoft wants to own that fix too.

Governor Newsom struck a deal giving every state agency, city, and county access to Claude at half price, plus free training. The DMV and the state's health agency are already using it. It's a bet that AI helps government workers move faster without replacing them, and a signal that public-sector AI adoption has moved past the pilot stage into policy.

SpaceX agreed to acquire Cursor's parent company, Anysphere, for $60 billion in stock, the largest startup acquisition on record. Cursor is software that writes and edits code alongside human developers, already used by roughly two-thirds of the Fortune 500. The deal folds a coding tool built jointly with SpaceX's AI arm straight into the rocket company.

Kimi K2.7 Code, built by China's Moonshot AI, is now available inside GitHub's coding assistant. It's "open-weight," meaning the underlying model is public and cheaper to run than the proprietary options. For any company watching AI costs, it's proof that credible alternatives to the big-name models are now one click away inside tools they already use.

The EU's Chips JU program and quantum computing company Pasqal launched Q-PLANET, a 50 million euro effort to build a proper factory pipeline for neutral atom quantum chips, the type of chip that traps individual atoms with lasers to run calculations regular computers can't touch. Thirty-seven partners across 12 countries are involved. Translation: Europe is trying to manufacture its own quantum hardware instead of importing it later.

1X's humanoid robot Neo has new hands with 25 points of articulation and built-in touch sensors, letting it sense grip force the way a human hand does instead of just clamping down blindly. It can already pour tea, zip a jacket, and pick a screw up off the floor, and the company says it can build 10,000 of these hands this year. Hands like this are the actual bottleneck standing between humanoid robots and doing real household or warehouse work.

Chevron signed a 20-year deal to build a dedicated natural gas power plant next to Microsoft's planned AI data center campus in Pecos, Texas. The setup is "behind the meter," meaning Chevron generates the electricity on-site instead of pulling it off the shared public grid, skipping the years-long wait most new data centers face for grid power. First power is expected in 2028.

Together AI, which rents out AI computing power built on open-source models instead of proprietary ones, closed an $800 million round at an $8.3 billion valuation. NVIDIA and Saudi Arabia's Aramco Ventures both invested. The company also locked in more than 500 megawatts of computing capacity, enough to power a mid-size city, a sign of where the smart money still thinks the real bottleneck is: not model quality, but raw compute.