Analysis
Grok 4.5 launched July 8 under the SpaceXAI brand -- the name xAI's models now carry following xAI's absorption into SpaceX earlier this year -- priced at $2 per million input tokens and $6 per million output tokens, roughly half what comparable frontier models from OpenAI and Anthropic charge. Elon Musk described the model's performance on X as "roughly comparable to Opus 4.7, but much faster," a direct challenge aimed at Anthropic's current flagship.
The pricing move lands squarely in the middle of an already-aggressive frontier-model price war: OpenAI's GPT-5.6 family launched this same week with its own tiered pricing designed to undercut Anthropic, and Anthropic itself extended free access to Claude Fable 5 rather than let a promotional window lapse into paid usage during the same stretch. Grok 4.5 at $2/$6 per million tokens now undercuts both GPT-5.6's Terra tier ($2.50/$15) and Sol tier ($5/$30) on a pure price basis.
Independent evaluator Artificial Analysis ranked Grok 4.5 fourth on its GDPval-AA v2 benchmark -- behind Claude models but ahead of most other frontier competitors -- while pricing it at $0.49 per completed task, roughly 90% cheaper than the higher-ranked models above it. That combination of near-frontier capability at a steep cost discount is the core of xAI's competitive pitch: it doesn't need to win every benchmark outright if it can win on cost-per-completed-task for enterprises running inference at scale.
“Grok 4.5 at $2/$6 per million tokens now undercuts both GPT-5.6's Terra tier ($2.50/$15) and Sol tier ($5/$30) on a pure price basis.”
The launch is also the first major SpaceXAI product release since SpaceX's $60 billion acquisition of Cursor and the broader SpaceX-xAI merger that created a combined entity valued at $1.77 trillion after last month's record-breaking IPO. That corporate structure is unusual in the industry -- no other frontier AI lab is housed inside a company whose primary business is rockets and satellites -- and it gives xAI access to a capital base and compute-infrastructure ambitions that dwarf most standalone AI labs.
For founders building on frontier-model APIs, Grok 4.5's pricing is another reason to stay model-agnostic: a near-frontier-capability model at a steep discount changes the calculus for any application where inference cost scales with usage volume, not just accuracy. For investors, the real signal is Musk's willingness to price aggressively below cost-parity with OpenAI and Anthropic immediately after a record IPO gave SpaceXAI the capital cushion to absorb thinner margins while it builds market share.
The bear case is that ranking fourth on GDPval-AA v2 means Grok 4.5 is still not the frontier-capability leader, and price alone rarely wins enterprise AI contracts where reliability and safety track record matter as much as cost. What to watch next: whether OpenAI or Anthropic respond with matching price cuts, and whether Grok 4.5's cost advantage actually converts into enterprise adoption outside of X and Tesla's own internal usage.