SpaceXAI launches Grok 4.5 at steep developer discount as compute tension grows
The enterprise software cycle has reached a familiar inflection, and this time the argument is price. SpaceXAI released Grok 4.5, its newest large language model, targeting coding and agentic work at $2 per million input tokens…
HONG KONG— July 8, 2026
The enterprise software cycle has reached a familiar inflection, and this time the argument is price. SpaceXAI released Grok 4.5, its newest large language model, targeting coding and agentic work at $2 per million input tokens and $6 per million output tokens. That undercuts Anthropic's Opus 4.8, which carries a $5 per million input and $25 per million output price tag.
A tool for developers, built alongside Cursor
Grok 4.5 is SpaceXAI's first major model release since the company went public and acquired Cursor, the AI coding startup. The model was trained in tandem with Cursor and is positioned as a professional productivity tool rather than a consumer chatbot. Elon Musk, writing on X, described it as "an Opus-class model, but faster, more token-efficient and lower cost," and the company published benchmark data showing it outperforms Anthropic's Opus 4.8 on several measures. Musk separately characterized the model as "maximally truth-seeking."
The model is available in Grok Build, in Cursor across all subscription tiers, and through the SpaceXAI developer console. It is not yet available in the European Union.
Competitive pricing in a crowded field
SpaceXAI claims Grok 4.5 beats some models from OpenAI and Anthropic on speed, price, and performance. The caveat matters: that comparison does not extend to the largest, latest releases from those competitors. Musk said the company expects to surpass those models soon. SpaceXAI positions Grok 4.5 as a business contender on cost, while acknowledging the performance gap against rivals' flagship offerings.
OpenAI's GPT 5.6 Luna carries a $1 per million input token price and $6 per million output token price, matching Grok 4.5 on output while undercutting it on input. The broader release of GPT 5.6 is expected this week, following a limited rollout at the Trump administration's request. OpenAI is also releasing a new generation of voice models alongside that launch.
The compute question underneath
Sector-wide, the race to release models is running into a resource constraint. Against the backdrop of accelerating release cycles, SpaceXAI's position as both a model developer and a compute lessor is becoming harder to hold. As its own training and inference needs grow, the company will face a choice between deploying that capacity for its own products or leasing it to rivals as a revenue stream. Grok 4.5 itself was trained on the same infrastructure SpaceXAI currently supplies to Anthropic and Google.
Related reading
Source · 來源