Google DeepMind · closed · 2025-12

Gemini 3 Flash

Item: Gemini 3 Flash
Rating: 86.1

#8 index 86.1

the take
The value pick of the Gemini line. Near-Pro reasoning, it actually beats the old Pro on SWE-bench, and it costs a rounding error. This is what I default to for volume.

benchmarks

78.0

90.4

—

95.2

LiveCodeBench (coding)

—

specs

Context: 1.048576M
Input: $0.5/M
Output: $3/M
Speed: 165.7 tok/s
Modality: text, image, audio, video

strengths

Excellent price/perf ($0.50/$3)
Beats Gemini 3 Pro on SWE-bench
Fast (~166 tok/s), 1M multimodal context

weaknesses

Lower hard-reasoning ceiling than Pro
Superseded by 3.5 Flash on agentic work
Text-only output

Sources: [1][2]

Back to leaderboard Battle Gemini 3 Flash