SpaceXAI to release Grok 4.5 tomorrow, claims Opus-level speed at lower cost

SpaceXAI will release Grok 4.5 on July 9, with Musk claiming the 1.5-trillion-parameter model matches Claude Opus on performance at lower cost.

SpaceXAI will open Grok 4.5 to the public on Wednesday, with Elon Musk claiming the 1.5-trillion-parameter model matches Anthropic's Claude Opus on performance while cutting inference cost and latency.

"Grok 4.5 is an Opus-level model, but it's faster, more token-efficient, and costs less to run," Musk said in a post on X. The model has been in private beta at SpaceX and Tesla for several weeks.

Built on a new V9 foundation with roughly 1.5 trillion parameters, Grok 4.5 is the largest model SpaceXAI has shipped. The Information reported the model is being compared internally to Claude Opus 4.8 and OpenAI's GPT-5.5. This is the first SpaceXAI model built jointly with Cursor, the AI coding startup SpaceX is acquiring for $60 billion.

The release intensifies competition among frontier labs as three major model events converge on the same week. Google DeepMind targets July 17 for Gemini 3.5 Pro — rebuilt from scratch with a 2-million-token context window — while DeepSeek plans to graduate its V4 family from preview on the same date. SpaceXAI's vertically integrated approach — controlling the compute cluster, the model, and the coding tool that feeds training data — has no direct precedent among frontier labs.

Access will follow SpaceXAI's tiered rollout pattern: Heavy subscribers first, then a wider opening to SuperGrok tiers, according to feature flags spotted in the Grok web client on July 6. The model also appeared in Grok Build CLI, the company's terminal coding agent, within the same flow that prompts users to upgrade to the Heavy plan.

The V9 architecture incorporates Cursor's developer-workflow data as supplemental training material added after the initial pre-training run. An xAI engineer acknowledged that supplemental inclusion is "not quite as good as having it in initial training," according to a person familiar with the matter. The next model in training is being built to incorporate Cursor data from the beginning of pre-training, which the company expects to produce stronger coding performance.

Competitive claims face verification gap

The Opus-level claim carries weight because it rests on internal evaluations at SpaceX and Tesla rather than public benchmarks. No independent scores — from SWE-bench, Humanity's Last Exam, or LMArena — have been published for Grok 4.5. The AI industry has documented cases where self-reported benchmark scores diverged from independently measured results by more than 20 percentage points, making external validation critical.

For context, Anthropic's Claude Opus 4.7 — the current leader on the Artificial Analysis Intelligence Index — costs roughly $25 per million output tokens via API. OpenAI's GPT-5.5 runs about $30 per million. SpaceXAI has not disclosed Grok 4.5's pricing, but Musk's claim of lower cost suggests the company may undercut both, leveraging SpaceX's compute infrastructure to reduce serving expenses.

The release tests whether the compute and capital gained through SpaceX's merger with xAI — announced in February — translate into a system that narrows the gap with Anthropic and OpenAI on reasoning and coding, not just on parameter count. SpaceXAI's ability to vertically integrate model development with Cursor's coding tools and Tesla's real-world data creates an advantage that pure-play AI labs lack. But without independent benchmarks, the performance claims remain unverified, and the market will need to see third-party results before re-rating any AI-related tickers.

This article is for informational purposes only and does not constitute investment advice.