Leaked Gemini 4 Pro specs claim 2M context, $2.25/M input — likely bogus

airesearch12 · x · 2026-09-18

An alleged spec sheet for Gemini 4 Pro is circulating: 2M token context window, 256K output limit, $2.25/1M input and $11.25/1M output tokens, plus leaked benchmark scores of 95.3% on Terminal-Bench 2.1, 88.7% on DeepSWE v1.1, 86.8% on OSWorld-2.0 and 94.7% on CharXiv Reasoning, reportedly ahead of GPT-6 Astra on several tests. Google has confirmed Gemini 4 is in its most ambitious pre-training run yet. But the sharer explicitly doubts it: "big if true but likely isn't" — treat as unverified rumor.

Original post →

More from Models

Models channel →