Feature
Guide
Is AI Making Your Organization More Productive or More Dependent?
Key Takeaways
- Grok 4.5 handles complex engineering and end-to-end app builds.
- SpaceXAI claims tasks resolve in under half the steps of peers.
- Available via Cursor, Grok Build and API at $2/$6 per million tokens.
SpaceXAI (formerly xAI) launched Grok 4.5 on Wednesday, a model built for coding, agentic tasks and knowledge work. It was trained across tens of thousands of NVIDIA GB300 GPUs and ships in Grok Build, Cursor and the xAI console.
According to xAI, Grok 4.5 delivers roughly twice the token efficiency of comparable models, solving tasks in under half the number of steps. The company priced the model at $2 per million input tokens and $6 per million output tokens.
Grok 4.5 is not yet available in the EU. xAI said EU availability is expected in mid-July 2026.
Where Grok 4.5 Works Best
SpaceXAI reportedly invested heavily in data filtering and curation — including deduplication, quality scoring and domain-focused selection — to ensure that the data mixture stayed high-coverage and high-signal.
The company claims gains in coding capabilities, saying the model is able to design end-to-end functional apps with minimal specification. One use case shared included a simulation of the universe and solar system.
The new model also excels at office work, according to SpaceXAI. It’s able to build complex Excel models that involve research from the web, multi-sheet formula use and stickies or notes left behind for future reference.
Grok 4.5 is also able to work with PowerPoint and Word. The example use case included an outline of a five-slide quarterly business review.
Grok 4.5 is now available in Grok Build, in Cursor on all plans and from the SpaceXAI console. The company is currently offering the use of Grok 4.5 for free for a limited time.
LLMs Hit Token, Context Complications
Anthropic's Claude Opus 4.6 supports 1M tokens via opt-in beta. OpenAI matched that capacity with GPT-5.4, adding native computer-use capabilities. Z.ai's open-source GLM-5.2 also reached the 1M-token mark.
For coding pipelines, larger context windows enable multi-step task execution across many turns without human intervention.
Despite advances, high GPU costs and context limitations strain AI budgets. Some organizations are turning to Agent Mesh architectures as a solution, systems that pass only the most relevant information to LLMs, reducing token consumption and containing costs.
Recent SpaceX News
SpaceX completed what may be the largest IPO in history on June 12, 2026, raising $85.7 billion on Nasdaq under "SPCX" and pushing its market cap above $2 trillion.
The S-1 disclosed $18.7 billion in 2025 revenue against a $4.9 billion net loss, with Starlink as the sole profitable unit. The listing followed SpaceX's February 2026 all-stock merger with xAI at a combined $1.25 trillion valuation.
On June 16, SpaceX confirmed a $60 billion all-stock acquisition of Anysphere, maker of AI coding tool Cursor, expected to close by end of September 2026.
SpaceX's longest-range AI bet arrived in late June with the launch of Starmind, an orbital AI compute constellation targeting up to one million satellite nodes.