Grok 4.6 Tops CursorBench with 6X Lower Cost Than Rivals
XFreeze · x · 2026-08-22
Grok 4.6 achieved the #1 spot on CursorBench 3.2 with a score of 70.8% and remarkable efficiency. Priced at $2.81 per task, it costs roughly 6X less than Fable 5 Max ($17.32) and nearly 3X less than Opus 5 Max ($8.23). This combination of top-tier intelligence and low compute cost makes it highly powerful for long-horizon agentic coding tasks.
Related event: Grok 4.6 Tops CursorBench at One-Sixth the Cost of Rivals(3 posts)→
More from Models
- Grok Bot Ran a 12-Hour Data Analysis Nonstop, Handling API Rate Limits Itself — Daniel_Farinax · 2026-10-16
- One year of Recursive Language Models: GPT-5-mini beats GPT-5 by 110% on 132k-token tasks, cheaper too — a1zhang · 2026-10-16
- Users Report Claude Models Suddenly More Compliant After Forced Compaction — RealSharpNinja · 2026-10-16
- Prediction: Opus 5.5-level models on laptops by 2027, backed by Qwen 27B data — cephaloform · 2026-10-16
- Users Say Krea 2 Output Degraded Sharply Since Krea 3 Beta Launched — szywis · 2026-10-16
- LLM-as-a-Verifier lifts Nemotron 3.5 Super by 9-10 points on agentic benchmarks — simonguozirui · 2026-10-16