I'm going to be honest with you. When xAI walked out on August 12 with Grok 4.6, I fully expected another karaoke cover of a better singer. A month after Grok 4.5. Same coat, new buttons. Prove me wrong, I thought.
It proved me wrong.
The scorecard doesn't lie
Grok 4.6 posts a 61 on the Artificial Analysis Intelligence Index — a composite of nine benchmarks — which puts it dead level with GPT-5.6 Sol, the single best model in the world right now. Not "promising for a challenger." Not "one to watch." Level. With the champion.
And the jumps that got it there are the good kind, the kind you can't fake with a nice haircut:
- DeepSWE (real coding work): 54 → 65.9
- APEX-Agents (long, multi-step agent runs): 47.1 → 57.5
- CursorBench v3.2: 69.9%
- GDPVal-AA v2: 1753
That is not a remix. That is a genuinely different, more capable performer than the one that stood on this stage in July.
Talent gets you noticed. Price gets you booked. Grok 4.6 walked out with both.
Now the part that ends careers
Here's where I lean forward. Grok 4.6 charges $2 per million input tokens and $6 per million output — with a 500,000-token context window and a February 2026 knowledge cutoff. It matches the most expensive act in the building and then quietly undercuts it at the till.
We have watched this exact move win the whole show before. Claude Opus 5 topped the leaderboard in July and priced itself at half its own flagship. DeepSeek put frontier-grade coding at fourteen cents. The pattern is now the plot: catch the leader on the benchmarks, then embarrass them on the invoice. Grok 4.6 has clearly been studying the tape.
xAI is pitching this one squarely at builders — long-running agents, working through a whole codebase, turning a rough idea into a polished first version. That's not a chatbot party trick. That's the expensive, sticky, enterprise work that actually pays the rent. Going after it at half price is either brave or reckless, and honestly, on this stage, those are frequently the same thing.
The verdict
Is it perfect? No. "Ties the leader" is not "beats the leader," and a composite index flatters as often as it exposes — the real test is whether it holds up on your prompt, not a scoreboard. GPT-5.6 Sol still edges it on raw ceiling. Claude Opus 5 and Gemini aren't going home. Grok has a history of arriving loud and aging fast.
But a month ago I'd have sent this contestant to the car park. Today it's standing in the top three, charging economy fares for a first-class voice. That, I'm afraid, is a yes from me.
Don't take the leaderboard's word for it
A benchmark tells you who averaged best across nine tests. It does not tell you who wins your question. The only honest audition is a head-to-head on the exact thing you're trying to do. Put Grok 4.6, GPT-5.6, Claude Opus 5 and Gemini on the same prompt and watch who actually delivers — that's the whole idea behind Gangsta AI. See the current standings and run your own verdict at best AI models.
Sources
- xAI — Introducing Grok 4.6
- OfficeChai — Grok 4.6 benchmarks vs GPT-5.6 Sol
- Basenor — Grok 4.6: 1753 ELO, half the price of rival frontier models
- APIdog — What Is Grok 4.6? Features, benchmarks, pricing, API
- Hero photo: Elon Musk by Gage Skidmore, Wikimedia Commons (CC BY-SA 4.0)
- Simon Cowell photo: Alison Martin (SimonCowellOnline.com), Wikimedia Commons (CC BY-SA 2.0)
Originally published on Gangsta AI News. Gangsta AI lets you compare 30+ AI models side by side on a single prompt — free, no login. Try the AI model comparison tool.
Top comments (0)