Google reported an 87% score for Gemini 4 Argon on the Embodied Reasoning (ERQA) benchmark in its October 13, 2026 launch narrative, describing the result as relying on pure reasoning. Exact methods and samples remain unqualified.
Limits: Provider-reported result in Google’s public narrative, not an independently verified benchmark-owner result or a reproduced run. The selected model identifier is gemini-4-argon; exact evaluated checkpoint, snapshot/build and configuration are unknown. Google describes the result as relying on pure reasoning. This qualitative phrase does not establish no tools, a token/reasoning budget, a guardrail setting or a matched evaluation protocol. Benchmark revision, exact physical/spatial task subset, samples, execution date, harness, prompts, sampling, tools, token budget, aggregation, pass-k and uncertainty remain unknown. No overall model rank, matched head-to-head, real-world robot capability, guarded-deployment equivalence or operational success is established.
Google
