SpaceXAI dropped Grok 4.5 on Wednesday, and Elon Musk’s selling point isn’t that it’s the smartest model in the room. It’s that you won’t go broke running it.
The model runs $2 per million input tokens and $6 per million output. Compare that to Claude Opus 4.8 at $5/$25 or GPT 5.6 Sol at $5/$30. Musk himself put it bluntly on X: Grok 4.5 is “roughly comparable to Opus 4.7, but much faster.” Opus 4.7 is Anthropic’s previous flagship — already superseded by 4.8 and then by Claude Fable 5.
So yeah. They’re not leading.
But here’s where it gets interesting. On SWE Bench Pro, which measures software engineering problem resolution, Grok 4.5 scored 64.7% — enough to beat GPT 5.5’s 58.6%. Opus 4.8 sits at 69.2%, and Fable 5 dominates at 80.4%. The DeepSWE 1.1 benchmark tells a similar story: Grok 4.5 at 53%, behind Opus 4.8’s 59% and Fable 5’s 70%.
The real edge isn’t benchmark scores, though. It’s efficiency. On SWE Bench Pro tasks, Grok 4.5 averaged 15,954 output tokens per job. Opus 4.8 burned through 67,020 for the same work. That’s a 4.2x gap. For teams running AI at scale, those numbers compound into real savings on top of the already lower per-token pricing.
The model also cranks out 80 tokens per second — firmly in fast-model territory. SpaceXAI trained it on tens of thousands of Nvidia GB300 GPUs inside the Colossus supercomputer, using developer session data from Cursor (the company SpaceX has a pending $60 billion deal to acquire).
Grok 4.5 isn’t available in the EU yet. That’s expected mid-July. For everyone else, it’s live via API, Hermes, and Grok build with half a million tokens of context.
The bottom line? If you need frontier performance, Claude Fable 5 leads every category SpaceXAI chose to publish. If you need to run lots of capable code edits without burning through your budget, Grok 4.5 makes a strong argument. Speed and cost over raw capability — that’s the bet.
