← Grok
Wes Roth
Grok 4.5 just COOKED Claude and OpenAI
⏱ 35 min video · 3 min read9 Jul 2026Worth watching
TL;DR
Grok 4.5 launched as the first model co-trained by xAI and Cursor, placing it in the same performance tier as Claude Opus and GPT-5.5 at significantly lower cost and faster speed. The creator demonstrates it building a 3D sailing game, an Elder Scrolls-style RPG, and a 50-block procedural city, then proposes a Fable 5 as architect / Grok 4.5 as builder workflow that produced a massive 3D city for about $8.
Key points
1
Grok 4.5 benchmarks sit between GPT-5.5 Extra High and Claude Opus on Deep SWE 1.0, Terminal Bench, and SWE Bench Pro, confirming it is a genuine Opus-class model.
2
This is the first model trained jointly by xAI and Cursor, giving it access to real-world end-to-end software development data that Cursor collected from professional engineers.
3
Grok 4.5 is substantially cheaper and faster than Claude Opus and GPT-5.5, making it better positioned on the intelligence-per-dollar curve for high-volume coding work.
4
A two-model workflow was demonstrated: Fable 5 acts as architect generating specs and quality checks, Grok 4.5 acts as construction crew executing them — a 50-district animated 3D city cost roughly $8 total vs an estimated $70-80 if Fable 5 had done everything.
5
The xAI and Cursor merger is strategically significant: Cursor previously paid API prices to frontier labs for their best models, but now has its own competitive model, reducing dependency on Anthropic and OpenAI.
Actionable insights
→
Use Grok 4.5 as your default coding workhorse for grunt-level implementation tasks where Opus-class intelligence is sufficient — it delivers comparable quality at lower cost and higher speed.
→
Adopt a tiered model workflow: use a frontier reasoning model like Fable 5 only for architecture, specs, and quality review, then hand execution off to Grok 4.5 to dramatically cut API spend.
→
Grok Build (xAI CLI coding agent, equivalent to Claude Code) now uses Grok 4.5 as its default model, making it worth evaluating as a direct Claude Code or Codex alternative for agentic coding pipelines.
Notable quotes
“Fable 5 is the architect, and Grok 4.5 is the construction crew.”
“Almost a megabyte of handwritten 3D code costs less than a burrito.”
“At some point don't you get tired of being wrong — just all the time — just how many bad takes in a row can you have.”
Worth watching?
✅
Worth watching the full video?
Watch if you want to see the actual gameplay and city demos in action — the key benchmarks, workflow strategy, and cost numbers are all captured here, but the visual proof of what Grok 4.5 built is more convincing in motion.
Saved you some time? The creator still deserves a like.
Found this by searching? That's the hard way.
summree watches the AI channels you follow and sends you the summary the moment a video drops. Nothing gets past you. From £4/month.
Try it free