- Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, compared to $5/$25 for Anthropic's Opus 4.8 [1] - The model is built on SpaceXAI's 1.5-trillion-parameter V9 foundation and was trained alongside Cursor, which SpaceX acquired for $60 billion in June 2026 [2] - On SWE-Bench Pro, Grok 4.5 scores 64.7% — behind Opus 4.8's 69.2% — but uses roughly 4.2x fewer output tokens per task [3] - Cursor acknowledged that an earlier snapshot of its codebase was accidentally included in training data, potentially inflating some benchmark results [3] - SPCX shares traded at $150.33 on July 9, up 1.4% from the prior close but down 12% from $171.57 at the start of the month
[4] SpaceXAI on July 8 released Grok 4.5, a coding- and agentic-work-focused model that Elon Musk described as "an Opus-class model, but faster, more token-efficient and lower cost" [1]. It is the company's first major model launch since SpaceX went public on the Nasdaq in June at a $1.77 trillion valuation and completed its $60 billion acquisition of Anysphere, the maker of the AI coding editor Cursor
.
[2]Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens in its standard tier, undercutting Anthropic's Opus 4.8 — which charges $5 and $25 respectively — by a wide margin [3]. A faster variant runs $4 input and $18 output. The model is available immediately through Grok Build, the Cursor editor on all plans, and the SpaceXAI developer console, though EU access is expected mid-July
.
[2]The launch lands one day before OpenAI's planned public release of GPT-5.6, setting up a direct head-to-head between the three leading foundation-model labs in the enterprise coding segment [2].
The Model #
Grok 4.5 is built on SpaceXAI's V9 foundation, a 1.5-trillion-parameter model trained across thousands of Nvidia GB300 GPUs with an emphasis on data quality and deduplication [2]. The company positioned it as a workhorse for coding, app-building, research, and routine knowledge work rather than a consumer chatbot
.
[1]SpaceXAI claims the model runs at approximately 80 tokens per second and uses roughly 4.2 times fewer output tokens than Opus 4.8 to solve equivalent SWE-Bench Pro tasks — an average of 15,954 tokens versus 67,020 [3]. That token efficiency, combined with lower per-token pricing, translates to dramatically lower per-task costs for developers.
Benchmark Performance #
The benchmark picture is mixed. On SWE-Bench Pro, the industry's most-watched coding evaluation, Grok 4.5 scores 64.7% — trailing Opus 4.8's 69.2% by 4.5 points but beating OpenAI's GPT-5.5 at 58.6% [3]. On Terminal-Bench 2.1, Grok 4.5 hits 83.3%, edging out Opus 4.8 (78.9%) while essentially matching GPT-5.5 (83.4%)
.
[3]On DeepSWE 1.0, Grok 4.5 scored 62.0% versus 55.8% for Opus 4.8 and 64.3% for GPT-5.5 [3]. The results suggest the model is competitive with, but not uniformly superior to, leading rivals — despite Musk's Opus-class framing.
A transparency issue complicates the picture: Cursor acknowledged that an earlier snapshot of its codebase was accidentally included in Grok 4.5's training data, which may have inflated results on Cursor's own internal benchmarks. The contaminated data has been removed for future model versions [3].
Pricing and Competitive Positioning #
The pricing strategy is the most aggressive element of the launch. At $2/$6 per million tokens, Grok 4.5 is 60% cheaper on input and 76% cheaper on output than Anthropic's Opus 4.8. It is roughly in line with OpenAI's GPT-5.6 Luna, priced at $1/$6 [2].
The pricing underscores SpaceXAI's strategy of leveraging its Cursor acquisition — for which it agreed to pay $60 billion in June — to build a vertically integrated coding platform [2]. By training the model on Cursor data and embedding it directly into the Cursor editor, SpaceXAI is positioning Grok 4.5 not just as an API product but as the default intelligence layer inside a widely used developer tool.
Stock and Market Context #
SpaceXAI shares (SPCX) traded at $150.33 on July 9, up 1.4% from the prior close of $148.30 [4]. The stock has been under pressure since early July, falling from $171.57 on July 1 — a decline of roughly 12% over eight trading sessions amid broader selling in high-multiple tech names
.
[4]SpaceX went public on June 12 at $135 per share, the largest IPO in history, and closed its first day at $160.95 — a 19% gain [5]. The company's market capitalization stands at approximately $1.97 trillion
.
[4]## What's Next The immediate test for Grok 4.5 will be developer adoption. SpaceXAI's integration with Cursor gives it a built-in distribution channel, but Anthropic and OpenAI both have deep enterprise relationships and their own coding-agent products.
OpenAI's GPT-5.6, delayed due to U.S. national security concerns over advanced AI misuse risks, is scheduled to go public on July 9 [2]. That release will provide the first direct, real-world comparison against Grok 4.5 in the same week — a rarity in a market where model launches have historically been staggered by months.
EU availability for Grok 4.5 is expected by mid-July [2].
Further sources #
[[1] Scoop: Musk's SpaceXAI releases new model, Grok 4.5 — Axios ↗](https://www.axios.com/2026/07/08/spacexai-grok-new-model)
[[2] SpaceXAI launches Grok 4.5 model for coding, agentic tasks — Yahoo Finance / U.… ↗](https://finance.yahoo.com/technology/ai/articles/spacexai-launches-grok-4-5-204651036.html)
[[3] SpaceXAI And Cursor Release Grok 4.5, Beats Opus 4.8 And GPT 5.5 On Some Benchm… ↗](https://officechai.com/ai/spacexai-and-cursor-release-grok-4-5-beats-opus-4-8-and-gpt-5-5-on-some-benchmarks/)
[[4] SPCX real-time quote and historical price data — Financial Modeling Prep ↗](https://financialmodelingprep.com)
[[5] SpaceX IPO takeaways: SPCX closes at $161, jumping 19% after record debut — CNBC ↗](https://www.cnbc.com/2026/06/12/spacex-ipo-spcx-live-updates.html)
The stories that matter, in one email. Free — unsubscribe anytime.