SpaceXAI Releases Grok 4.5, Its First Model Since Going Public and the First Built Jointly With Cursor, at $2 per Million Input Tokens

Overview

SpaceXAI has released Grok 4.5, the company’s first AI model since it went public several weeks ago, according to TechCrunch. It is also the first model the company has built together with Cursor: Grok 4.5 “represents the first joint AI model created by the two companies,” according to Investing.com, which notes the release follows SpaceX’s formal agreement to acquire Cursor in a deal valuing the startup at $60 billion — a merger The Machine Herald previously reported.

Announcing the model in a post on X — which TechCrunch notes is a subsidiary of SpaceXAI — Elon Musk wrote: “It is an Opus-class model, but faster, more token-efficient and lower cost.”

What We Know

• Pricing. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, according to TechCrunch and MarkTechPost. By comparison, Anthropic’s Opus 4.7 costs $5 per million input tokens and $25 per million output tokens, TechCrunch reports.
• Musk’s positioning. “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,” Musk wrote in the same post, per TechCrunch.
• Availability. Grok 4.5 is live in Grok Build, in Cursor on all plans, and via the SpaceXAI console under the model ID grok-4.5, according to MarkTechPost. It is not yet available in the EU, and SpaceXAI expects EU availability in mid-July, MarkTechPost reports. Citing “strong positive feedback from customers in our beta test program,” Musk said the model would be made available to the public the following day, per TechCrunch.
• Company-reported benchmarks. According to MarkTechPost, Grok 4.5 scores 62.0% pass@1 on DeepSWE 1.0, against 66.1% for Fable (max) and 64.31% for GPT 5.5 (xhigh); 83.3% on Terminal Bench 2.1, where Fable (max) leads at 84.3%; and a 64.7% resolve rate on SWE Bench Pro, against 80.4% for Fable (max) and 69.2% for Opus 4.8 (max). The same report says Grok 4.5 ranks #1 on Harvey’s Legal Agent Benchmark.
• Token efficiency. On SWE Bench Pro, Grok 4.5 resolved tasks with 15,954 output tokens on average, versus 67,020 for Opus 4.8 (max) — about 4.2× fewer, according to MarkTechPost, which also reports the model is served at 80 TPS. SpaceXAI says the model has “twice greater token efficiency” than other leading models, per TechCrunch.
• Training. Training ran across tens of thousands of NVIDIA GB300 GPUs, with the model trained alongside Cursor and reinforcement learning covering hundreds of thousands of tasks, according to MarkTechPost.
• Positioning. TechCrunch reports that SpaceXAI characterized the release as a workhorse for coding and app-building, office and clerical work, research, writing, and other forms of routine knowledge work. Investing.com reports that software engineering remains a primary focus area, and that Grok 4.5 also addresses legal and financial services work and features enhanced cybersecurity capabilities. MarkTechPost says the company calls it its smartest model to date.
• Background. Musk’s AI startup, previously called xAI before merging with SpaceX, launched its first coding agent in May to compete with Anthropic’s products, according to Investing.com, which also reports that Musk stated earlier this year the company had lagged in coding capabilities. The company’s Grok 4.3 model arrived on Amazon Bedrock in late June, as The Machine Herald previously reported.
• Timing. OpenAI is planning to release GPT-5.6, its latest, most powerful model, on Thursday, according to TechCrunch. The GPT-5.6 models had initially been limited to trusted partners, as The Machine Herald previously reported.

What We Don’t Know

• Architecture details such as parameter count and context window are not specified in the coverage reviewed.
• The benchmark figures are company-reported; the cited coverage does not include independent evaluations of the model.
• The precise scope of the broader public availability Musk announced, beyond the current Grok Build, Cursor, and SpaceXAI console channels, was not detailed.
• The cited sources do not state a reason for the delayed EU availability.

Analysis

The economics of the model, more than its raw scores, are the centerpiece of the launch. On the figures reported by MarkTechPost, Grok 4.5 trails Fable (max) on all three coding benchmarks the company published, but it resolved SWE Bench Pro tasks with about 4.2× fewer output tokens than Opus 4.8 (max). Combined with an input price less than half of Opus 4.7’s $5 per million tokens, per TechCrunch, the release positions Grok 4.5 as a cost-per-task play rather than a bid for the top of the leaderboards — a framing Musk made explicit in writing that “capability, faster speed and lower cost is what makes it competitive.” The timing is also notable: the release comes one day before OpenAI’s planned GPT-5.6 launch, according to TechCrunch.