July 9, 2026
Elon Musk’s SpaceXAI officially launched its most powerful flagship model to date, Grok 4.5, marking the company’s first major product release since its public debut several weeks ago . Musk took to his social media platform, X, to personally announce the public availability of the AI, which he unequivocally described as an “Opus-class model” — a direct reference to Anthropic’s prestigious line of large language models designed for the most complex and intensive tasks .
However, in a move that signals a significant shift in the AI industry’s competitive landscape, Musk was quick to emphasize that Grok 4.5 is not just about raw power; it is fundamentally designed to be much faster, dramatically more token-efficient, and offered at a fraction of the cost of its leading competitors . This strategic combination of capability, speed, and affordability appears to be the core of SpaceXAI’s new strategy to carve out a dominant space in an increasingly crowded market, directly challenging incumbents like OpenAI and Anthropic .
The development of Grok 4.5 represents a massive technological and financial undertaking. The model was forged using tens of thousands of NVIDIA GB300 GPUs in an ultra-large-scale training operation, a testament to the immense computational firepower Musk’s companies can marshal . A critical piece of this puzzle was the landmark collaboration with the AI programming startup Cursor, which SpaceXAI recently agreed to acquire in a staggering $60 billion deal .
This partnership was not just a business transaction but a foundational element of the model’s training. By incorporating trillions of tokens of Cursor’s proprietary data—which captures how real-world developers interact with codebases, tools, and intelligent agents—Grok 4.5 learned not just the syntax of code, but the dynamic process of collaboration between humans and AI in software engineering . This focus makes Grok 4.5 the first model from SpaceXAI specifically optimized for coding and complex, long-running agentic tasks, a capability Musk previously admitted his AI division was lagging in, which had prompted a significant internal restructuring to address the shortfall .
The performance metrics released alongside the model showcase its highly competitive position within the top tier of AI models. On several core engineering benchmarks, Grok 4.5 demonstrated its prowess by achieving a 64.7% score on SWE Bench Pro, directly surpassing Opus 4.7 and approaching the performance of even more advanced models like Opus 4.8 . Furthermore, on the more challenging DeepSWE 1.0 benchmark, Grok 4.5 scored 62.0%, surpassing Opus 4.8 and closely trailing GPT-5.5 . While Musk himself was frank in his internal assessment, stating that Grok 4.5’s overall capabilities are roughly comparable to Anthropic’s Opus 4.7, he stressed that the model’s true advantage lies not in being the absolute best in class, but in delivering this Opus-level performance with unprecedented speed and efficiency . This “fast-follower” strategy allows it to compete on a different vector, prioritizing cost-effectiveness and operational velocity for users.
Where Grok 4.5 truly disrupts the market is in its cost and efficiency. The model boasts an inference speed of up to 80 TPS (Tokens per second), which is officially noted as being “faster than flash-class models” . This speed is complemented by a radical improvement in token efficiency, which is a key driver of operational costs for AI applications. The model uses just over 15,000 output tokens to solve problems on the SWE Bench Pro, a task that requires over 67,000 tokens for Opus 4.8—a 4.2x reduction in token consumption . This efficiency is directly reflected in its aggressive pricing strategy.
Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens for its standard version, with a faster premium tier offered at $4 for input and $18 for output . This pricing is a stark contrast to its rivals. For comparison, Opus 4.7 costs $5 per million input tokens and a hefty $25 per million output tokens . Even OpenAI’s tiered models, with their most expensive Sol version at $5/$30, are significantly undercut by SpaceXAI’s offering, making top-tier AI intelligence accessible at a fraction of the cost .
The launch of Grok 4.5 is more than just a product announcement; it is a strategic power play during a pivotal week for the AI industry. It comes amidst what many are calling a “model war,” occurring just days before OpenAI was set to release its powerful GPT-5.6, which had previously faced governmental delays . By launching first with a compelling “Opus-level performance for less” narrative, SpaceXAI is positioning itself as a formidable alternative.
The model’s capabilities extend beyond programming, with demonstrated proficiency in creating complex Excel models, generating PowerPoint presentations, and crafting coherent Word documents, showcasing its ambition to be an enterprise workhorse . Looking ahead, Musk hinted at an even more significant leap with the next version, promising to integrate learnings from real-world engineering challenges at Tesla, SpaceX, and other ventures to create a “data-training-application” loop that competitors cannot easily replicate . For now, Grok 4.5 is immediately available to the public in the SpaceXAI console, Grok Build, and through Cursor, marking a new era in which high-performance, affordable AI is becoming the new standard .
