Google

Gemini 3.0 Flash

Released 5 months ago · Dec 17 2025

Benchmarks

15 tracked

Agentic coding

SWE-Bench Pro

49.6%

Coding

SWE-Bench Verified

78%

Agentic terminal coding

Terminal-Bench 2.1

58%

Multi-step tool use

MCP Atlas

62%

General tool use

Toolathlon

49.4%

Multidisciplinary reasoning

Humanity's Last Exam

33.7%

no tools

Abstract reasoning

ARC-AGI-2

33.6%

Science

GPQA Diamond

90.4%

Agentic computer use

OSWorld-Verified

65.1%

Agentic financial analysis

Finance Agent v2

42.6%

Knowledge work

GDPval-AA

1204

Chart reasoning

CharXiv Reasoning

80.3%

Multimodal reasoning

MMMU-Pro

81.2%

Spatial reasoning

Blueprint-Bench 2

Long context

MRCR v2 (8-needle)

67.2%

128k average

22.1%

1M pointwise

About Gemini 3.0 Flash

Gemini 3.0 Flash is an AI model released by Google on Dec 17 2025, 27 days after Gemini 3.0 Pro Image. Benchmark results (shown above) cover SWE-Bench Pro, SWE-Bench Verified, Terminal-Bench 2.1, MCP Atlas, Toolathlon, Humanity's Last Exam, and 9 more.

About Google: Builds the Gemini family of models through Google DeepMind. Integrates AI across Google products.