gpt.college

Google · Gemini 3.7 · Reviewed 2026-09-02

Gemini 3.7 Flash

Google's generally available Flash reasoning model with native text, image, audio, video, and file input.

All models
Overall · 0–100—
Status
active
Released
2026-08-13
Context
1,048,576 tokens
Parameters
Undisclosed
Inputs
Text · Image · Audio · Video · File
Type
multimodal
Max output
65,536 tokens
API model ID
gemini-3.7-flash
Reasoning
low / medium / high
Knowledge cutoff
2026-03
Training cutoff
—
Weights / license
Unavailable
Price / 1M tokens
$0.75 in · $3.75 out
Cached input
$0.075
Broad native multimodal input1M-token contextConfigurable low, medium, and high reasoning
CodingReasoningResearchMultimodalAgentic

Independent benchmark scores

No snapshot score

Coding

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Reasoning

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Research

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Multimodal

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Agentic

—

No comparable score for this task lens in the selected independent snapshot.

Raw benchmark results

Source-backed rows with the original evaluation context preserved.

21 results
BenchmarkScoreSourceEvaluation notes
Agents' Last ExamEvaluated 2026-08-13 26.3% Google Vendor-reported Agents' Last Exam result.
Artificial Analysis Intelligence IndexChecked 2026-09-05 51 Artificial Analysis Artificial Analysis Intelligence Index v4.x; reasoning=low.
Artificial Analysis Intelligence IndexChecked 2026-09-05 53 Artificial Analysis Artificial Analysis Intelligence Index v4.x; reasoning=medium.
Artificial Analysis Intelligence IndexChecked 2026-09-05 56 Artificial Analysis Artificial Analysis Intelligence Index v4.x; reasoning=high.
AutomationBenchEvaluated 2026-08-13 30.4% Google Vendor-reported AutomationBench result; version and harness are unspecified in the supplied model card.
BioMysteryBench, human-difficultEvaluated 2026-08-13 43.5% Google Vendor-reported BioMysteryBench result; task set=human-difficult.
BioMysteryBench, human-solvableEvaluated 2026-08-13 87.1% Google Vendor-reported BioMysteryBench result; task set=human-solvable.
CharXiv Reasoning, no toolsEvaluated 2026-08-13 84.5% Google Vendor-reported CharXiv Reasoning result; tools=none.
CharXiv Reasoning + ToolsEvaluated 2026-08-13 88.7% Google Vendor-reported CharXiv Reasoning result; tools=enabled.
Code Arena WebDevEvaluated 2026-08-13 1588 Google Vendor-reported Code Arena WebDev score; score unit is Elo.
DeepSWE v1.1Evaluated 2026-08-13 65.3% Google Vendor-reported; benchmark=DeepSWE v1.1.
FrontierCode 1.1 MainEvaluated 2026-08-13 43.6% Google Vendor-reported Gemini 3.7 Flash Model Card result; benchmark=FrontierCode 1.1 Main.
GDM-MRCR v2Evaluated 2026-08-13 97% Google Vendor-reported GDM-MRCR v2 result; context setting=128K.
GDP.pdfEvaluated 2026-08-13 34% Google Vendor-reported GDP.pdf result.
GDPval-AA v2Evaluated 2026-08-13 1525 Google Vendor-reported GDPval-AA v2 score; score unit is Elo.
HLE-VerifiedEvaluated 2026-08-13 53.6% Google Vendor-reported HLE-Verified result.
LABBench2Evaluated 2026-08-13 82.1% Google Vendor-reported LABBench2 result.
LVBenchEvaluated 2026-08-13 85.4% Google Vendor-reported LVBench result.
OSWorld 2.0Evaluated 2026-08-13 47.9% Google Vendor-reported OSWorld 2.0 result; task release and harness should remain attached to the source context.
Terminal-Bench 2.1Evaluated 2026-08-13 85.8% Google Vendor-reported; benchmark version=Terminal-Bench 2.1; harness details are not supplied in the model card.
Terminal-Bench 3.0Evaluated 2026-08-13 14.9% Google Vendor-reported; benchmark version=Terminal-Bench 3.0; do not compare with Terminal-Bench 2.1 or 4.0.

Strengths

  • Broad native multimodal input
  • 1M-token context
  • Configurable low, medium, and high reasoning

Limitations

  • Parameter count and weights are not publicly disclosed
  • Benchmark scores depend on reasoning configuration

Model evidence · official · Checked 2026-09-02

Gemini 3.7 Flash model card

Official Google model page for Gemini 3.7 Flash limits, modalities, reasoning levels, release status, and model-card benchmarks.

Task lenses preserve raw scores. Radar axes use score / current leader × 100; overall is the equal-weight average of all five indices. Missing evidence is not a zero or a low rank.