GadgetBond

  • Latest
  • Features
  • How-to
  • Tech
    • AI
      • Apple Intelligence
      • Gemini AI
      • Google DeepMind
      • Anthropic
      • Claude AI
      • Claude Code
      • OpenAI
      • ChatGPT
      • Codex
      • Perplexity
      • SpaceXAI
      • Grok AI
      • Microsoft Copilot
      • Meta AI
    • Amazon
    • Apple
    • CES
    • Computing
    • Creators
    • Google
    • Meta
    • Microsoft
    • Mobile
    • NVIDIA
    • Samsung
    • Security
    • Smart Home
    • Sony
    • Xbox
    • YouTube
  • Transportation
    • Audi
    • BMW
    • Cadillac
    • E-Bike
    • Ferrari
    • Ford
    • Honda Prelude
    • Lamborghini
    • McLaren
    • Mercedes
    • Porsche
    • Rivian
    • Tesla
  • Culture
    • Gaming
    • Streaming
    • Apple TV
    • Disney
    • Hulu
    • Marvel
    • HBO Max
    • Netflix
    • Paramount
    • SHOWTIME
    • Spotify
    • Star Wars
Add GadgetBond as a preferred source to see more of our stories on Google.
Font ResizerAa
GadgetBondGadgetBond
  • Latest
  • Tech
  • AI
  • Features
  • Deals
  • How-to
  • Apps
  • Computing
  • Gaming
  • Mobile
  • Streaming
  • Transportation
Search
  • Latest
  • Deals
  • Buying Guide
  • How-to
  • Features
  • Tech
    • Amazon
    • Apple
    • CES
    • Computing
    • Creators
    • Google
    • Meta
    • Microsoft
    • Mobile
    • NVIDIA
    • Samsung
    • Security
    • Smart Home
    • Sony
    • Xbox
    • YouTube
  • AI
    • Apple Intelligence
    • Gemini AI
    • Google DeepMind
    • Anthropic
    • Claude AI
    • Claude Code
    • OpenAI
    • ChatGPT
    • Codex
    • Perplexity
    • SpaceXAI
    • Grok AI
    • Microsoft Copilot
    • Meta AI
  • Transportation
    • Audi
    • BMW
    • Cadillac
    • E-Bike
    • Ferrari
    • Ford
    • Lamborghini
    • McLaren
    • Mercedes
    • Porsche
    • Rivian
    • Tesla
  • Culture
    • Gaming
    • Streaming
    • Apple TV
    • Disney
    • Hulu
    • Marvel
    • HBO Max
    • Netflix
    • Paramount
    • SHOWTIME
    • Spotify
    • Star Wars
Follow US
AISpaceXTech

SpaceXAI and Cursor launch Grok 4.5 for real-world software work

SpaceXAI and Cursor trained Grok 4.5 on real developer usage patterns, hoping to produce a model that behaves well in day-to-day engineering.

By
Shubham Sawarkar
Shubham Sawarkar's avatar
ByShubham Sawarkar
Editor-in-Chief
Shubham Sawarkar is a technology journalist at GadgetBond, covering Apple, AI, consumer technology, software, gadgets, and product launches. His work explores emerging technologies, industry developments, product...
Follow:
- Editor-in-Chief
Jul 9, 2026, 9:00 AM EDT
Share
GadgetBond is reader-supported. We may earn a commission from purchases through our links. Learn more
SpaceXAI logo displayed in white against a bright blue sky with large white clouds, featuring the SpaceX-inspired wordmark representing the merged SpaceX and xAI brand.
Image: SpaceXAI
SHARE

SpaceXAI has launched Grok 4.5, a new flagship model designed less as a chatbot for casual questions and more as a working engine for software development, long-running AI agents, and the kind of document-heavy knowledge work that fills real offices. The company calls it its smartest model so far, and the pitch is straightforward: frontier-level coding ability without the sluggishness or eye-watering token bills often attached to the most capable AI systems.

That claim arrives at a moment when the AI race is increasingly about whether models can actually complete work, rather than merely produce impressive-looking answers. Grok 4.5 is built around that shift. It is intended to navigate codebases, use tools over many steps, fix bugs, build applications, research information, and help create spreadsheets, Word documents, and PowerPoint presentations.

The headline detail is its development alongside Cursor, the widely used AI coding platform. Cursor says Grok 4.5 is a mixture-of-experts model jointly trained with SpaceXAI using trillions of tokens drawn from Cursor data – a potentially important distinction because that data reflects the messy, iterative reality of developers and coding agents working through projects, rather than just static code found online.

For years, the demo standard for AI coding has been deceptively simple: ask for a small website, a game, or a script and see what comes back. The harder test is whether a model can enter an unfamiliar repository, understand what is broken, make changes across files, run commands, recover from errors, and explain its decisions. That is the territory SpaceXAI is chasing with Grok 4.5.

The company says it trained the model on coding, science, engineering, and mathematics data, then used reinforcement learning across hundreds of thousands of tasks focused on multi-step software engineering and technical work. Training ran across tens of thousands of NVIDIA GB300 GPUs, according to SpaceXAI, which also says it put heavy emphasis on filtering, deduplicating, and curating data rather than simply throwing more tokens at the model.

Its benchmark results are promising, though they deserve the usual dose of skepticism. SpaceXAI reports that Grok 4.5 achieved a 29% pass-at-one score on SWE Marathon, ahead of the listed scores for Opus 4.8 and Fable, while reaching 83.3% on Terminal Bench 2.1 – essentially tied with the strongest competing models cited by the company. On DeepSWE 1.0, however, its 62% score trails Fable’s 66.1% and GPT-5.5‘s 64.31%.

That uneven scoreboard is actually more revealing than a clean sweep would be. There is no single benchmark that settles which model is “best,” because coding tasks vary wildly: some reward careful planning, some reward command-line execution, some reward persistence, and some reward a model’s ability to stay focused through a sprawling multi-file repair. Grok 4.5 looks positioned near the front of that pack rather than unambiguously ahead of it.

Where SpaceXAI is pressing hardest is efficiency. It says Grok 4.5 runs at 80 tokens per second and used an average of 15,954 output tokens on SWE-Bench Pro tasks, compared with 67,020 for Claude Opus 4.8 Max in its comparison – about 4.2 times fewer tokens. If those results consistently transfer to day-to-day development work, they matter. Faster completions and shorter reasoning traces can make agentic workflows feel more usable, while also lowering the price of letting an AI take several attempts at a difficult task.

The pricing is equally aggressive for a model being marketed at the top end of the market: $2 per million input tokens and $6 per million output tokens. Grok 4.5 is available through SpaceXAI’s developer console, Grok Build, and Cursor, with the company offering limited free use in Grok Build and Cursor at launch.

For developers, the Cursor integration may be the most practical part of the launch. A highly capable model only becomes valuable when it sits inside the tools people already use, can inspect files and terminal output, and has enough context to follow work over time. SpaceXAI is trying to turn Grok 4.5 into that kind of collaborator – one that can move from a vague request to a functioning application, rather than stopping at a code snippet and a cheerful explanation.

The company is also trying to broaden the idea of what “AI coding” means. In Grok Build, it says the model can assemble complex Excel workbooks that include web research, multi-sheet formulas, and notes for future users. It is also being positioned for PowerPoint and Word workflows, including slide diagrams built with native PowerPoint shapes and more polished document drafting.

That office-work angle is not just a side feature. The most commercially useful AI assistants will likely be the ones that can cross boundaries between research, analysis, code, spreadsheets, and presentations without forcing users to stitch together half a dozen products. A developer might ask an agent to investigate a bug, read error logs, patch the software, summarize the root cause, and create a short update for the rest of the team. Grok 4.5 is being sold as a model for exactly that chain of work.

There are, naturally, caveats. SpaceXAI’s speed, token-efficiency, and benchmark comparisons are company-reported figures, and real-world performance depends heavily on the task, prompt, tool setup, and human oversight. An AI agent that confidently modifies a production codebase still needs guardrails, tests, code review, and someone accountable for the final decision. Benchmark leadership is useful evidence, but it is not a substitute for that discipline.

Availability is also not fully global from day one. Grok 4.5 has launched in Grok Build, Cursor, and the SpaceXAI console, but SpaceXAI says it is not yet available in the European Union and expects EU access in mid-July.

Still, the broader signal from this release is clear. SpaceXAI is no longer framing Grok primarily as a personality-driven assistant connected to a social platform. With Grok 4.5, it is aiming at the more consequential – and lucrative – market for AI systems that can carry out technical work. The question now is not whether Grok 4.5 can produce a slick demo from one prompt. It is whether developers and businesses will trust it enough to make it part of the daily machinery of getting things done.

Your browser does not support the video tag.

Discover more from GadgetBond

Subscribe to get the latest posts sent to your email.

Topic:Grok AISpaceXAIxAI
1 Comment
  • Marin T. Kael's avatar Marin T. Kael says:
    Jul 11, 2026, 2:44 PM EDT at

    the thing i'd want to see is per task-type behavior, not one aggregate score. real usage is bimodal and a single mean washes that out. did they break the eval down by workflow, or is it just a headline number?

    Reply as
    Reply on the Fediverse

    Remote Reply

    Original Comment URL

    Paste the comment URL into the search field of your favorite open social app or platform.

    Your Profile

    Or, if you know your own profile, we can start things that way!
    Why do I need to enter my profile?

    This site is part of the ⁂ open social web, a network of interconnected social platforms (like Mastodon, Pixelfed, Friendica, and others). Unlike centralized social media, your account lives on a platform of your choice, and you can interact with people across different platforms.

    By entering your profile, we can send you to your account where you can complete this action.

Leave a Reply Cancel reply

Most Popular

Apple launches Health AI Usage Study with OpenEvidence to evaluate conversational health tools
Get 25% off the dive-rated Garmin fēnix 8 AMOLED smartwatch with this Early Black Friday deal
Garmin Instinct 3 AMOLED gets a $200 Black Friday discount
Google Sheets now lets you publish a Sheets canvas as a standalone link
Google Chat now lets you schedule meetings with a single click using Gemini

Also Read

Illustration showing Apple Support Community members represented by colorful Memoji avatars alongside icons for Apple services and products, including Photos, Home, iCloud, Apple Music, Accessibility, Mac, iPhone, and Settings.

Apple is shutting down its user-to-user support forums after nearly 30 years

A gray rounded-square icon featuring a white hard drive with five vertical vent slots on its front edge, centered against a light blue background.

Apple splits macOS Device Control into two privacy settings

Robert Pattinson poses as James Bond in a black tuxedo, aiming a pistol toward the camera against a dark background, with the 007 logo in the lower-left corner.

Robert Pattinson is officially the next James Bond as Amazon MGM Studios announces new 007

White pill reading "Personal Intelligence" below a Gemini sparkle and beside a Google Wallet icon, surrounded by faded Google app logos.

Gemini can now analyze your spending with Google Wallet and Plaid integration

Three Apple HomeView smart displays in different colors and configurations, showcasing the 7-inch screens with a smart home dashboard, FaceTime video call, and personalized photo display.

Apple HomeView goes on sale today for $299 with a 7-inch display and Siri AI

Models wearing Google's new smart glasses collections designed with Warby Parker and Gentle Monster.

Google opens pre-orders for new Gemini-powered smart glasses from Warby Parker and Gentle Monster

Google's smart glasses setup screen displaying responsible-use guidelines, including restrictions against recording without permission, filming in sensitive spaces, and covering or damaging the external recording indicator light.

Google outlines privacy safeguards for its upcoming AI glasses, including anti-tampering recording lights

Google Gemini 4 Hydrogen logo displayed against a blue gradient background, featuring the multicolored Gemini star icon and white product name.

Google launches Gemini 4 Hydrogen for faster coding and everyday AI workflows

Company Info
  • Homepage
  • Support my work
  • Latest stories
  • Company updates
  • GDB Recommends
  • Daily newsletters
  • About us
  • Contact us
  • Write for GadgetBond
  • Editorial guidelines
Legal
  • Privacy Policy
  • Cookies Policy
  • Terms & Conditions
  • DMCA
  • Disclaimer
  • Accessibility Policy
  • Security Policy
  • Do Not Sell or Share My Personal Information
Socials
Follow US

Disclosure: We love the products we feature and hope you’ll love them too. If you purchase through a link on our site, we may receive compensation at no additional cost to you. Read our ethics statement. Please note that pricing and availability are subject to change.

Copyright © 2026 GadgetBond. All Rights Reserved. Use of this site constitutes acceptance of our Terms of Use and Privacy Policy | Do Not Sell/Share My Personal Information.