xAI released Grok 4.6 today, and it went straight onto the bench: graded the same day, as a successor to Grok 4.5, on the Week 33 suites the rest of the field just ran. These are its day-one numbers, not the vendor's.
What the results show
- The upgrade shows when the ask is vague. Handed a thin issue report with little to go on, Grok 4.6 now works out and fixes the real problem as well as anything on the bench - tied with Claude Fable 5 for the lead, where Grok 4.5 sat mid-pack. That jump alone lifts it to second on the overall board.
- Everyday engineering work looks unchanged. On well-specified, realistic tasks it performs like its predecessor: solid, a small step behind the leaders.
- You pay for the improvement in quota and wait. A graded pass burns about twice the share of a SuperGrok week that Grok 4.5 did, and tasks take roughly twice as long.
The bottom line: Grok 4.6 is clearly the stronger model, but the strength comes at the cost of the one edge its predecessor had. Fast and good enough for everyday work was Grok 4.5's niche; twice the wait and twice the burn puts Grok 4.6 in a straight fight with the top tier - one it would be a surprise to see it win outside benchmarks.
The succession is stitched as one line on Trends, so the step shows in context. Grades are comparable within the same suite version - explore the complete Week 33 results.