Logic of Logic
thursday, august 6, 2026 · the day's ai, attributed published by trilot llc · wyoming
brief productsbusiness

SpaceXAI ships Grok 4.5, built with Cursor

Grok 4.5 launched on a 1.5T-parameter base trained partly on real Cursor coding sessions, priced at $2/$6 per million tokens, well under Opus and GPT-5.6.

xAI, which now brands itself SpaceXAI after SpaceX’s stock-for-stock acquisition earlier this year, released Grok 4.5 on July 8. The model runs on a new 1.5-trillion-parameter foundation, internally called V9, and was trained in part on real developer sessions from Cursor, SpaceX agreed to acquire Cursor’s parent, Anysphere, in a $60 billion deal in June.

The base model is priced at $2/M tokens in and $6/M tokens out, with a faster variant running $4/$18. That undercuts Anthropic’s Opus 4.8 ($5/$25) and lands below OpenAI’s cheapest GPT-5.6 tier, Luna ($1/$6), while sitting well under Sol’s $5/$30. xAI says Grok 4.5 runs at roughly 80 tokens per second and uses about a quarter of the output tokens Opus 4.8 needs on comparable SWE-Bench Pro tasks.

On xAI’s own published benchmarks, Grok 4.5 beats Opus 4.8 on two of four (DeepSWE 1.0 and Terminal-Bench 2.1) and loses on two others; no independent third-party benchmark has run yet. It’s available now inside Grok Build, across every Cursor plan, and via the SpaceXAI console, with a European rollout expected by mid-July.

For operators comparing coding-agent costs, Grok 4.5’s token efficiency plus its price point make it a reasonable test case against Opus 4.8 or GLM-5.2 on your own workload before committing budget, especially given the benchmarks are vendor-published and unverified so far. See our coverage of Anthropic’s Fable 5 pricing shift for how the frontier-pricing picture is moving on the other side.

sources 2 cited
1 x.ai Introducing Grok 4.5 2 techcrunch.com SpaceXAI releases Grok 4.5, which Elon describes as an 'Opus-class model'
next