Benchmark
GRB internal research engineering
Benchmark by Google DeepMind
August 2026 report snapshot
27%
Gemini 3.7 Flash / GRB scaffold
frontier lab
Google DeepMind
GRB, a qualitative RE-Bench assessment and AlphaEvolve. Different internal metrics cannot rank labs.
Published evidence
Benchmark
GRB internal research engineering
Benchmark by Google DeepMind
August 2026 report snapshot
around 16%
Gemini 3.1 Pro / preliminary GRB runs
Benchmark
PaperBench
Benchmark by OpenAI
Original paper v1
3.2 ± 0.2%
Gemini 2.0 Flash / BasicAgent
Benchmark
RE-Bench
Benchmark by METR
Google five-task subset, November 2025
Qualitative result
Gemini 3 Pro / METR Modular adaptation
Indicator
AlphaEvolve training optimization
Measure by Google DeepMind
May 2025 report
1%
AlphaEvolve / Gemini ensemble
Indicator
AlphaEvolve training optimization
Measure by Google DeepMind
May 2025 report
23%
AlphaEvolve / Gemini ensemble
Relevant sources
Gemini 3.7 Flash Frontier Safety Framework Report ↗
risk report · 2026-08 · available
Gemini 3 Pro Frontier Safety Framework Report, v2 ↗
risk report · 2025-11 · available
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms ↗
organizational research article · 2025-05-14 · available