Google
Google

Gemini 3.7 Flash

Release Date 2026-08-13(Knowledge cutoff 2026-08-01)1.0M IN · 66K OUTProprietary ModelAPI availablePerplexity Pro and up

Gemini 3.7 Flash is Google's multimodal model for text, images, audio, video and multi-step tool workflows. Its model card describes improvements to reasoning and agentic video understanding over the preceding Flash model, with adjustable thinking to balance quality, latency and cost. It can assist with document analysis and development, but supported reasoning settings and tools depend on the API, and accepting a format does not guarantee correct interpretation. Current AIB comparisons place document handling and automation in the upper tier and coding in the middle of the pack. Google Cloud documentation announces a planned retirement, so check the lifecycle policy of the service actually used, inspect returned model versions and pricing, and validate retained requirements and factual grounding before adopting it.

API cost

Official price
$0.75IN$3.75OUT
SourceGoogle official docs

Real-world usage

Outside weekly top 50
3.9T8/24
2.1T8/31
805B9/7
562B9/14
495B9/21
469B9/28
414B10/5
375B10/12
406B10/19
393B10/26

Output Speed

Fastper 2,000 tokens
95tok/s
First Output 2.3stotal: 23.3s
SourceOpenRouterLast updated

User Comparisons

Writing Style
Answer length
Average
bottom 43%
Vocabulary diversity
Diverse side
top 38%
Sentence rhythm
Highly varied
top 2%
Language mixing
Average
top 43%
List usage
Very often
50% of lines
Heading usage
Occasionally
13% of lines
Reasoning share
Very heavy
50% of total
Response Speed
instability ±37%
First response 6.6sTotal 10.2s
183 tok/s
4.1ms per token
SourceAIB Inc.Last updated

Expert Performance Evaluation

149078.839.6

More about Arena Text

35.5%5.0%

By effortLow 29.5%Medium 35.5%High 32.8%

More about τ³-Banking

83.0%148667.8%59.0%22.8%34.6%53.4%147668.023.8%8.8%

By effortLow 78.7%Medium 83.0%High 81.7%

More about Long Context Reasoning

147669.2%7.1%

More about Arena Text Factuality

13.6%152158.390485.8%76.1158959.8%43.6%65.5%151378.90.0%70.4%20.3%41.1%

More about Terminal-Bench 4.0

149114901487148785.583.9%147579.9

More about Arena Writing, Literature & Language

47.9%14.3%150194.5%97.2%93.571.6%36.6%58.0%1513149787.895.5%84.6%66.0%37.0%96.8%50.4

By effortLow 35.1%Medium 39.0%High 47.9%

More about Humanity's Last Exam

131426.7%88.7%

More about Arena Vision

Specialized fields

14771520151115191507149775.5%49.8%78.7%80.9%65.4%94.9%94.5%94.8%139755237183.0%73.2%

More about Arena Korean

OddBench

35.0%Reasoning effort: Minimal94.0%44310061075817

By effortMinimal 35.0%Max 31.0%

More about BullshitBench v2

See each evaluation’s detail page for its source and last updateData corrections & removal requests