Now that Grok 4.5 is available, we’d love to hear how it’s working for you.
Some things we’re especially curious about:
What are you using it for beyond coding? This is the first model we’ve built for more than software engineering. If you’ve pointed it at data science, research, finance, legal work, or anything else you do on a computer, we want to hear how it went.
How does it handle long-running, multistep tasks? Grok 4.5 was trained to investigate problems, use tools creatively, recover from mistakes, and verify results before calling it done. Does that match your experience?
How does it compare to Composer and other frontier models in your day-to-day? Anything noticeably better (or noticeably worse)? It should be solving tasks in notably fewer steps; let us know if you’re seeing that.
This is also our first model trained jointly with SpaceXAI, and the kind of multistep, tool-heavy work it’s built for is exactly what’s hardest to capture in a benchmark, so your feedback shapes where the model goes next.
Drop your thoughts below (and leave a Request ID if you have one)! We read everything.
Under the command of GPT-5.5, XHigh launched a team of subagents, with Grok 4.5 High as the primary one. Refactoring involves downsizing the project’s Python files due to the LOC-per-file size limit.
What surprised me most was that it runs scripts directly in the console to extract entire methods into separate files and then fixes any issues.
In one call Grok rewrote ~6k lines of code, making edits via edit_tool in less than a hundred lines. However, he did a lot of console magic, so maybe he partially wrote the code through the shell
It feels like Next Composer, not Next Grok, and I’m VERY happy about that.
Upd: The Verifier Subagent forced him to write more than a hundred lines to make everything work
Upd2: Organized the file from 2.2k to 0.24k LOC with that tool
It’s really strange—the model was released just 2.5 hours ago, and I’ve already hit the usage limits, which aren’t even visible yet. I guess it works like ChatGPT or Claude, with limits resetting every 3–5 hours?
To be 100% clear on this.
Is Grok 4.5 included in the ‘Auto + Composer’ limit? I have plenty of that limit and still get the ‘Switched to composer-2.5 after reaching API limit.’ message.
Come on, my friend, I can see exactly how my workflow is going. Yes, I offloaded most of my tasks to Grok 4.5 to see what kind of results I’d get. But what bothered me is the lack of a transparent limit tracking system, like the ones in OpenAI’s ChatGPT and Anthropic’s Claude. It feels like they haven’t been completely honest with us about the model’s limits.
