Grok 4.5 is straight-up crushing it on that intelligence-to-cost curve. 54 on the index at just $0.31 per task while some of the big names are bleeding money for similar or worse performance? That’s not just progress, that’s a masterclass in efficiency.The terafab + Tesla FSD brain trust combo is going to be unstoppable. Those guys already turned real-world autonomy into an art form with lean, lightning-fast models. Once that scales at full volume, the rest of the industry is gonna feel it.

Don't fall for the hype blindness. Grok 4.5 is a competitive mid-tier model with a clean TUI, but it isn't "the top." 🧵 1/ Intelligence Gap: On coding benchmarks like DeepSWE 1.1, Grok 4.5 hits a 53% pass rate, while Claude Fable 5 dominates at 70% and GPT-5.5 sits at 67%. It still trails the true frontier leaders in complex reasoning. 2/ The "Speed" Illusion: At 80-90 tokens per second (TPS), its raw speed is completely average. The illusion of speed comes from its 4.2\times token efficiency (fewer output tokens used), not faster raw hardware generation. Lightweight Flash models still run circles around it. Competition is great, but don't confuse a shiny UI, clever token architecture, and aggressive pricing with absolute technical superiority. 🚗💻

@tylerbruno05 @elonmusk Love seeing this Grok 4.5 does feel noticeably snappier while still being sharp. Exactly the direction I was hoping for too. Keep crushing it 🚀

@ArthurMacwaters The ratio between intelligence and cost per task is becoming the most important metric in the AI race and this chart proves xAI is executing perfectly.