Leaked Benchmarks Hint Grok-4 Tops Key AI Reasoning Tests

Some leaked Grok-4 and Grok-4 Code results - 35% on HLE, 45% with reasoning!! - 87-88% on GPQA - 72-75% on SWE Bench (Grok 4 Code)

BREAKING: Grok 4 benchmarks leaked Impressive numbers: 🔹 grok-4-0629 Standard: HLE 35, GPQA 87, AIME’25 95 Test Time: HLE 45, GPQA 88, AIME’25 — 🔹 grok-4-code-0629 Standard: SWEBench 72, Terminal Bench — Test Time: SWEBench 75, Terminal Bench

BREAKING 🚨: Grok 4 will be SOTA 🤖 - 35% on HLE, 45% with reasoning - 87-88% on GPQA - 72-75% on SWE Bench (for Grok 4 Code) * Not official benchmarks, I plotted these based on the leaked scores and results from the web for other models

Grok-4 and Grok-4 Code benchmarks are 🔥🔥 - 35% on Human last Exam benchmark and 45% with reasoning. - GPQA : 87-88% - SWE Grok 4 Code : 72-75% N Note : These are leaked numbers we have to wait for final. h/t : legit_api

Grok-4 and Grok-4 Code on benchmarks - 35% on HLE, 45% with reasoning!! - 87-88% on GPQA - 72-75% on SWE Bench (Grok 4 Code)