SpaceXAI made Grok 4.5 available to the public Thursday — its first model since Elon Musk’s company acquired coding startup Cursor — and it led with the price. Grok 4.5 runs $2 per million input tokens and $6 per million output. Anthropic’s Claude Opus 4.8, the model Musk named as the bar to beat, costs $5 and $25. That is roughly a quarter of the output price for something SpaceXAI insists is “Opus-class”.
Musk’s framing, posted as the model went live: “an Opus-class model, but faster, more token-efficient and lower cost.” A chart shipped with the announcement claims Grok 4.5 edges Opus 4.8 on several coding and agentic benchmarks. This is a developer tool, not a chatbot play — it’s live in Grok Build, inside Cursor on every plan, and through the SpaceXAI console, though not yet in the EU. It is also the payoff on the Cursor deal: SpaceXAI trained the model on Cursor’s coding data, then pointed it straight at Cursor’s users.
The launch completed a first. The same day, OpenAI threw open GPT-5.6 to everyone and Anthropic’s Sonnet 5 sat as the default model — three frontier labs shipping public models in a single stretch, all cutting price at once. But Grok 4.5 arrived with the loudest claim and the thinnest paper trail. As of launch there is no full system card and no independent third-party benchmarks — only SpaceXAI’s own scorecard. “Maximally truth-seeking,” Musk called the model. The one thing you cannot yet check is whether the benchmarks are.
Our take: The price is the real news, and the price is verifiable. $2 in and $6 out is a genuine shot at Anthropic’s flagship economics, and it will drag the whole coding-model tier down with it. The performance claim is not verifiable — not yet. A vendor chart is marketing; “Opus-class” means whatever the maker says it means until independent evals weigh in. Treat the price as a reason to test Grok 4.5 this week and the benchmark chart as a hypothesis. The pattern from the GPT-5.6 launch and Sonnet 5 holds: the model keeps getting cheaper, and the moat keeps sliding to the workflow you build on top of it.
What to watch
- Independent benchmarks. SpaceXAI published its own numbers and no system card. Wait for third-party evals — SWE-bench, Terminal-Bench, deployment-style tests — before you re-platform production work onto Grok 4.5.
- The price response. $2/$6 undercuts Opus 4.8 by more than half on input and roughly three-quarters on output. Watch whether Anthropic, OpenAI, and Google answer on price — or on capability.
- Cursor as a channel. Grok 4.5 is native in Cursor on every plan. The acquisition turned a popular coding tool into a distribution pipe; that’s the strategic story, not the token rate.
- The EU gap. Grok 4.5 isn’t live in the EU at launch. Watch how long that lasts and what it says about where the model can be shipped.