Grok 4.5 Drops Same Day as GPT-5.6, Pricing War Erupts at $2/Million Tokens 🪙

SpaceXAI released Grok 4.5 to the public on Wednesday, marking the first model from the company since the SpaceX-xAI merger closed in February and SpaceX's pending $60 billion deal to acquire Cursor. Elon Musk announced the rollout on X, stating that SpaceXAI will make Grok 4.5 available to the public based on strong positive feedback from customers in the beta test program. "It is an Opus-class model, but faster, more token-efficient, and lower cost," Musk wrote.

Grok 4.5 runs on xAI's 1.5 trillion-parameter V9 foundation, with Cursor coding data added in supplemental training. The model entered private beta at SpaceX and Tesla on June 28, and SpaceXAI trained it in collaboration with Cursor AI on tens of thousands of Nvidia GB300 GPUs inside Colossus, the Memphis supercomputer with total capacity across more than 200,000 GPUs. Pricing is set at $2 per million input tokens and $6 per million output, undercutting Claude Opus 4.8 at $5 input and $25 output, as well as OpenAI's GPT 5.6 Sol at $5 input and $30 output.

OpenAI confirmed Wednesday that GPT-5.6 Sol, Terra, and Luna will launch publicly on Thursday. "We're expanding preview access globally now," the company posted on X. OpenAI previewed the GPT-5.6 family in late June, limiting initial access to a small group of vetted partners, and has since received Commerce Department approval for the broad launch.

On benchmarks published at launch, Grok 4.5 scored 53% on DeepSWE 1.1, behind Claude Opus 4.8 at 59% and GPT 5.5 at 67%, with Claude Fable 5 topping the chart at 70%. On SWE Bench Pro, Grok 4.5 posted 64.7%, beating GPT 5.5's 58.6% but trailing Opus 4.8's 69.2% and Fable 5's 80.4%. SpaceXAI's benchmarks compare against GPT 5.5 because GPT 5.6 launched hours after Grok 4.5. Musk clarified Grok 4.5's positioning in a separate post, calling it "roughly comparable to Opus 4.7, but much faster" and framing speed and cost over raw capability as a deliberate tradeoff.

Token-efficiency data underscores the pricing argument: on SWE Bench Pro tasks, Grok 4.5 used an average of 15,954 output tokens per job, compared with 67,020 for Opus 4.8, a 4.2x gap. The release revives a long rivalry between Musk and OpenAI, which he co-founded in 2015 and left in 2018; Musk sued OpenAI and Sam Altman in 2024 alleging breach of an early nonprofit pledge, though a jury dismissed the lawsuit in May as untimely. With Anthropic's Fable 5 already available after the recent lifting of a U.S. export control ban, all three labs now have their latest flagships in public circulation.