Benchmark

Google — reported benchmarks

Google · source page ↗ · last checked Aug 22, 2026, 12:03 PM

Reported benchmarks · Gemini 3.1 Pro

captured Aug 22, 2026, 12:03 PM
BenchmarkScore
GPQA Diamond94.3% · Graduate-level science
ARC-AGI-277.1% · Abstract reasoning
SWE-Bench Verified80.6% · Agentic code evaluation
Humanity's Last Exam44.4% · No tools
LiveCodeBench Pro2887 Elo · Competitive programming
Terminal-Bench 2.068.5% · Terminus-2 harness
SciCode59.0% · Scientific research coding
BrowseComp85.9% · Agentic task
MMMLU92.6% · Multilingual
MCP Atlas69.2% · Tool use coordination
APEX-Agents33.5% · Agentic multi-step tasks

Vendor-reported via automated web search — not independently verified. See the cited matrix on /models.

In the news · Google

Importance-filtered press coverage (Google News) mentioning Google. Headlines link to the original; verify before acting.

Change history