Overview
<cite index="3-1">Elon Musk's SpaceXAI launched Grok 4.5, its smartest model yet, and its first release since going public and acquiring the artificial intelligence (AI) coding startup Cursor.</cite> <cite index="1-5,1-6">Grok 4.5 is SpaceXAI's strongest model built to excel at coding, agentic tasks, and knowledge work.</cite>
Positioning and Pricing
<cite index="3-5">"It is an Opus-class model, but faster, more token-efficient and lower cost," wrote Musk in a post on X, referencing what was until recently Anthropic's top model family.</cite> However, Musk also clarified the model's precise standing: <cite index="9-7">"Our internal assessment is that Grok 4.5 is roughly comparable to Opus 4.7, but much faster. The combination of capability, faster speed and lower cost is what makes it competitive."</cite>
The pricing differential is central to SpaceXAI's go-to-market strategy. <cite index="5-1,5-2">Grok 4.5 is available on the xAI Application Programming Interface (API), priced at $2 per million input tokens and $6 per million output tokens, with configurable reasoning effort.</cite> <cite index="6-11">Rival Anthropic's Claude (claude) Opus 4.8 is priced at $5 per million input tokens and $25 per million output tokens.</cite> <cite index="20-17,20-18">That is roughly 4.2x more expensive on output cost alone.</cite>
Training and Infrastructure
<cite index="1-24">Grok 4.5 was trained across tens of thousands of NVIDIA Graphics Processing Units (GPUs) GB300s, with training and stability techniques designed for large-scale runs.</cite> <cite index="14-5">The model is built on the V9 architecture at approximately 1.5 trillion parameters.</cite> <cite index="16-25">Grok 4.5 was trained on developer session data from Cursor (cursor-anysphere), including debugging traces and real code edits rather than static repositories.</cite> <cite index="6-17">SpaceX had announced it would buy Anysphere, the startup behind Cursor, in an all-stock deal worth $60 billion to boost its presence in the enterprise AI tools market.</cite>
<cite index="7-9">Anthropic pays SpaceX $1.25 billion a month for compute access through its Colossus data centers, and Google pays $920 million a month for the same kind of access.</cite> <cite index="3-13,3-14">This model was trained using the same compute capacity SpaceXAI is leasing to its competitors Anthropic and Google — and as its own compute needs grow, SpaceXAI may have to choose between using capacity for its own models or leasing it to others as a revenue stream.</cite>
Benchmark Performance
SpaceXAI published four coding benchmarks alongside the launch; the results are mixed relative to Anthropic's current flagship. <cite index="1-20,1-21">On the SWE Marathon resolution rate, Grok 4.5 scored 29.0%, ahead of Opus 4.8 at 26.0%. On Terminal Bench 2.1, Grok 4.5 scored 83.3%, behind only Claude Fable 5 and GPT 5.5.</cite> <cite index="16-29,16-30,16-31,16-32">On DeepSWE 1.1, Grok 4.5 scored 53%, behind Claude Opus 4.8 at 59% and Fable 5 at 70%. On SWE Bench Pro, Grok 4.5 posted 64.7%, while Opus 4.8 leads at 69.2% and Fable 5 sits at 80.4%.</cite> <cite index="14-31">With only three of nine major benchmarks published and no third-party evaluations at launch, Grok 4.5 results remain vendor-reported.</cite>
Availability and Market Context
<cite index="1-11">Grok 4.5 is available in Grok Build, in Cursor (cursor-anysphere) on all plans, and from the SpaceXAI console.</cite> <cite index="1-13,1-14">The model is not yet available in the European Union; EU availability is expected in mid-July.</cite>
<cite index="4-12">Spending data from Ramp cited by CNBC showed Cursor's share of the AI coding category falling from 41% in June 2025 to about 26% by May 2026, while Anthropic came to control roughly half the market.</cite> <cite index="4-14,4-15">Grok 4.5 is engineered to close that gap — not by out-thinking Claude, but by underpricing it. If the model delivers most of the frontier's capability at a fraction of the cost per task, price-sensitive enterprise workloads may migrate, pressuring incumbents on their most profitable API traffic.</cite>
<cite index="1-36,1-37">Grok 4.5 is served at fast-model speeds of 80 tokens per second (TPS) and, combined with what SpaceXAI claims is twice greater token efficiency than comparable leading models, is designed to deliver results more quickly and at lower costs.</cite> Independent verification of those efficiency claims had not been published at time of writing.