xAI, now SpaceXAI, ships Grok 4.5 with near-Opus coding on a 1.5T-param model

Grok 4.5 entered private beta at SpaceX and Tesla built on a 1.5-trillion-parameter V9 model, with xAI (now officially SpaceXAI after February's acquisition) claiming near-Opus coding and agentic performance and touting a data-integration partnership with Cursor. Snorkel's independent testing framed it against real professional work, and Forbes covered the launch as the company's strongest model yet. The company committed to a monthly training-release cadence for the rest of 2026.

A July 12 Grok Build changelog fleshed out the developer story: new authentication-pinning options, per-skill token-cost display, improved clipboard handling, and subagent tool enhancements — incremental but signaling xAI is serious about the coding-agent surface.

Community debate skipped the benchmarks and went straight to politics. Elon Musk's post declaring Grok 'the most politically neutral and objectively truth-seeking AI' drew 20K likes but also renewed worries about systematic skew; opinions split between 'more politically correct than GPT' and fears of ideological tuning. Pricing drew broad praise — roughly $2/$6 per million tokens and a claimed '4x better reasoning efficiency compared to Opus.'

A separate HN thread dissecting what Grok's build CLI transmits back to xAI (406 points, 157 comments) fed telemetry and privacy concerns — a recurring anxiety as agentic CLIs proliferate. The open question is whether SpaceXAI's monthly cadence can sustain quality gains, or whether it becomes another entry in the industry's model-version fatigue.