name = "Gemini 3.7 Flash"
description = "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning"
family = "gemini-flash"
release_date = "2026-08-13"
last_updated = "2026-08-13"
attachment = true
reasoning = true
temperature = true
tool_call = true
structured_output = true
knowledge = "2026-03"
open_weights = false

[limit]
context = 1_048_576
output = 65_536

[modalities]
input = ["text", "image", "video", "audio", "pdf"]
output = ["text"]

[[benchmarks]]
name = "FrontierCode"
score = 43.6
metric = "score"
version = "1.1 Main"
source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
date = "2026-08-13"

[[benchmarks]]
name = "DeepSWE"
score = 65.3
metric = "resolve rate"
version = "1.1"
source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
date = "2026-08-13"

[[benchmarks]]
name = "Terminal-Bench"
score = 85.8
metric = "accuracy"
version = "2.1"
source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
date = "2026-08-13"

[[benchmarks]]
name = "AutomationBench"
score = 30.4
metric = "accuracy"
dataset = "private set"
source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
date = "2026-08-13"

[[benchmarks]]
name = "GDP.pdf"
score = 34.0
metric = "accuracy"
source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
date = "2026-08-13"

[[benchmarks]]
name = "GDM-MRCR"
score = 97.0
metric = "accuracy"
variant = "128k average, 8-needle"
version = "v2"
source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
date = "2026-08-13"