SpaceXAI shipped Grok 4.5 on July 8 at $2 per million input tokens and $6 per million output, and roughly eighteen hours later OpenAI opened GPT-5.6 Sol, Terra, and Luna to the public. Five frontier-class APIs are now live simultaneously, and the pricing spread between them is the actual story.

Grok 4.5 is SpaceXAI's first flagship co-developed with Cursor, trained on tens of thousands of NVIDIA GB300 GPUs. The company claims it's "roughly comparable to Opus 4.7, but much faster," with "twice greater token efficiency" than leading models. Elon, whose Space Exploration Technologies conglomerate now houses the AI unit, called it an "Opus-class model, but faster, more token-efficient and lower cost." TechCrunch's benchmark table read the same way most Grok benchmark tables read: competitive with the frontier, just short of best-in-class. A faster premium tier runs $4/$18. EU availability is slated for mid-July.

The pricing math is what changed overnight. Anthropic's Opus 4.7 sits at $5/$25. Grok 4.5's standard tier is $2/$6. That's not a discount, it's a repricing of what Opus-class inference is allowed to cost.

OpenAI's release the next morning tightened the vise. GPT-5.6 Sol lists at $5/$30, with a Sol Ultra variant above it. Terra, positioned at half the cost of GPT-5.5-equivalent performance, comes in at $2.50/$15. Luna, the small model, is $1/$6. OpenAI announced on X Tuesday evening that it was "expanding preview access globally," and by Wednesday the tiers were live.

The gating is worth noting. Per Engadget, the Trump administration granted OpenAI permission for wider release only after additional testing by the Department of Commerce's Center for AI Standards and Innovation, with company technical staff dispatched to DC to address concerns in person. The process traces to a June AI cybersecurity executive order asking major developers to voluntarily submit leading models for government review, roughly a 30-day window. In a late-June statement, OpenAI said it doesn't believe "this kind of government access process should become the long-term default."

That last detail is the one that'll age. A pre-release federal review, for a commercial API, from a company on record opposing the review's institutionalization, shipping the same week as an unregulated competitor from Musk's orbit and ahead of Anthropic Mythos. The frontier didn't just become a five-model market this week. It became a market where the price sheet, the regulator, and the political sponsor of each lab are all suddenly legible in the same frame.

Sources

https://x.ai/news/grok-4-5
https://techcrunch.com/2026/07/08/spacexai-releases-grok-4-5-which-elon-describes-as-an-opus-class-model/
https://tech.yahoo.com/ai/articles/spacex-spcx-launches-grok-4-071421210.html
https://www.nextgov.com/artificial-intelligence/2026/07/openais-advanced-gpt-56-models-be-available-public/414651/
https://www.engadget.com/2210308/openai-rolls-out-gpt5-6-july-9/