Inference
EN
Google

Gemini 3.8 Flash

Gemini 3.8 Flash is a reasoning model from Google. 66 benchmarks count toward its index, in 7 categories.

  • Closed weights
  • Reasoning
  • Input: text · image · video · file · audio
  • Released 2026-09
RankedEnough results for an index.
IndexAll categories, out of 100.67.8±2.5Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.68/sMedian 67/s
First tokenSeconds to the first output token.3.0 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.825batch $0.375Median $0.60
Output / 1MUS dollars per 1M output tokens.$4.13batch $1.88 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1500±4 #9Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
google/gemini-3.8-flash
Max outputMaximum output tokens per request.
66K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.083
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
3 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
Google AI StudioMaker$0.75$3.751.05M
Google Vertex AIMaker$0.75$3.751.05M
OpenRouterRouter$0.75$3.751.05M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.24 counted65.343 / 218
CodingCode writing and repair.16 counted66.235 / 209
ReasoningLogic problems and puzzles.10 counted68.328 / 227
MultimodalTasks with images and text.4 counted67.119 / 84
KnowledgeFacts and expert knowledge.6 counted70.917 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted91.3—
MathMath problems.5 counted64.334 / 168

Results

66 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
ARC-AGI-1 (semi-private)Reasoning98.5%1 / 126
Artificial Analysis GPQA DiamondKnowledge95.3%2 / 199
LiveBench Instruction FollowingInstructionAbove fitFar above this model's index.81.4%1 / 66
LiveCodeBenchCoding89.5% ±0.93 / 142
GPQA diamond · Epoch AIKnowledge95.4% ±1.45 / 217
GPQA Diamond · Vals AIKnowledge94.4% ±1.54 / 135
Artificial Analysis MMMU-ProMultimodal85.6%4 / 113
MMLU ProKnowledge90.2% ±0.35 / 135
MMMU ProMultimodal89.1% ±0.84 / 92
Terminal-Bench 2.1 · BenchLMAgentic89.4%5 / 86
ARC-AGI-2 (semi-private)Reasoning89.2%11 / 131
Artificial Analysis Humanity's Last ExamKnowledge47.8%20 / 222
LiveBench LanguageKnowledge87.8%6 / 66
Terminal-Bench 2.1 · Vals AIAgentic81.3% ±0.48 / 75
SimpleQA VerifiedKnowledge69.7% ±1.510 / 91
OTIS Mock AIME 2024-2025Math98.9% ±0.922 / 200
Artificial Analysis Omniscience AccuracyKnowledge50.5%19 / 164
Artificial Analysis ITBench-AAAgentic52.5%3 / 24
Chess PuzzlesReasoning58.0% ±5.014 / 90
Artificial Analysis Long Context ReasoningReasoning81.3%34 / 217
Critical Physics TasksReasoning18.3%34 / 214
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal71.2%7 / 39
Mystery Game PuzzlesReasoning47.0% ±5.016 / 80
Artificial Analysis SciCodeCoding56.6%23 / 114
DeepSWEAgentic73.8%12 / 59
LiveBench ReasoningReasoning89.3%18 / 80
CharXiv Reasoning without toolsMultimodal86.2%4 / 17
GDPval-AA normalizedAgentic46.8%34 / 133
SWE-benchCoding80.0% ±1.824 / 87
Vibe Code Bench v1.1Coding78.7% ±3.934 / 123
cursorBench32Coding69.2%5 / 18
Artificial Analysis Terminal-Bench v2.1Agentic87.6%6 / 21
MazeBench: visual spatial reasoning in a 3D open worldAgentic4.0%14 / 48
ARC-AGI-3 (semi-private)Reasoning10.4%8 / 26
ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable jobAgentic24.0%6 / 18
SkillsBenchCoding58.0% ±4.412 / 35
FrontierMath-Tiers-1-3-v2-PrivateMath68.4% ±2.832 / 90
OSWorld 2.0Agentic59.0%9 / 25
LiveBench MathematicsMath91.6%25 / 66
LiveBench ResearchAgentic79.1%20 / 52
Artificial Analysis Tau3-BankingAgentic44.9%9 / 22
IOICoding56.9% ±2.319 / 45
Code MigrationCoding36.5% ±4.235 / 77
LiveBench Agentic CodingAgentic54.2%32 / 66
LiveBench Browser UseAgentic64.0%25 / 49
Agent Arena task outcomeAgentic1.329 / 54
Vibe Code Bench 1-100Coding18.8% ±3.314 / 26
ScarfBench: Cross-Framework Application Migration in Enterprise JavaCoding66.0%6 / 11
Terminal-Bench 4.0Agentic19.7% ±2.334 / 61
LiveBench Coding 2.0Coding82.2%29 / 52
Hermes Index: agentic capability and cost per task inside Hermes AgentAgentic38.1%9 / 16
Artificial Analysis AutomationBenchAgentic59.9%13 / 23
Artificial Analysis AnalystAgentAgentic46.3%15 / 26
LiveBench Long AgenticAgentic30.1%31 / 51
Agent Arena steerabilityAgentic−1.333 / 54
MirrorCodeCoding38.9% ±11.924 / 39
ProofBench v1.1Math48.0% ±5.035 / 54
FrontierMath-Tier-4-v2-PrivateMath22.0% ±6.548 / 72
LiveBench DesignAgentic72.7%35 / 52
Artificial Analysis GDP.pdfAgentic21.0%15 / 22
Artificial Analysis CWE-Bench-AAAgentic52.5%13 / 19
Agent Arena command recoveryAgentic−0.735 / 50
Furniture AssemblyReasoning31.7% ±5.927 / 37
Artificial Analysis CyberGym-E2E-AAAgentic8.4%14 / 19
Terminal-Bench 4.0.0Agentic19.1% ±1.717 / 23
Terminal-Bench 4.1.0Agentic18.2% ±1.717 / 23
Artificial Analysis DeepsecBench-AAAgentic14.5%16 / 21
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal80.7%20 / 26
Artificial Analysis Cyber IndexAgentic25.1%15 / 19
cursorBench40Coding39.6%17 / 21
LiveBench CodingCoding72.5%54 / 66
Artificial Analysis Terminal-Bench v4.0Agentic19.7%18 / 22
FrontierSWE v2Coding19.6%18 / 22
Multimodal Knowledge Agent BenchmarkAgentic35.3%16 / 19
Bug Hunt BenchCoding18.0%23 / 25
LiveBench Data AnalysisReasoning54.0%76 / 80
Medical Long Context Reasoning (MLCR-AA)Reasoning10.6%22 / 22
Finance Agent (v2)Finance61.4% ±0.12 / 79
LegalBenchLegal87.0% ±0.47 / 149
ARC-AGI-2 (public eval)Reasoning92.5%11 / 122
ARC-AGI-1 (public eval)Reasoning98.0%11 / 116
EMBFinance72.2% ±2.48 / 75
ProgramBenchCoding1.0% ±0.716 / 65
TaxEval v2Finance74.4% ±0.836 / 143
Tax Agent BenchFinance32.1% ±3.423 / 82
MedCodeHealthcare48.1% ±2.231 / 106
MysteryMechanismScience36.5% ±3.29 / 30
SAGEEducation88.2% ±1.79 / 29
MedScribeHealthcare84.5% ±1.934 / 109
SRE BenchOther11.1% ±1.913 / 36
Legal Research BenchLegal38.9% ±3.438 / 90
Public Benefits Bench v1.1Public benefits65.3% ±1.222 / 50
MortgageTaxFinance65.3% ±0.944 / 97
Terminal-Bench-ScienceScience8.6% ±3.420 / 44
Time Horizon Index: KSPOther18.8%6 / 13
HLAB-Vals-1.1Legal12.1% ±2.618 / 35
FrontierMath-ErdosMath0.0%8 / 15
BioMysteryBenchScience62.2% ±1.725 / 30
CyberBench v1.1Other43.8% ±2.243 / 48
CUA-benchOther4.2%9 / 9
BioMysteryBench Human SolvableKnowledge88.8%—
LVBenchMultimodal87.1%—
LABBench2: An Improved Benchmark for AI Systems Performing Biology ResearchKnowledge86.2%—
Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindAgentic62.2%—
BioMysteryBench Human DifficultKnowledge56.5%—
HLE-VerifiedKnowledge54.9%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYARC Prize FoundationTerms restrict reuseVals AINo licence statedLiveBenchNo licence statedTerminal-BenchNo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index70.3 → 67.8
23 Oct 202612 Nov 2026
Output / 1M$4.13 → $4.13
23 Oct 202612 Nov 2026

Questions

How much does Gemini 3.8 Flash cost?
$0.825 per 1M input tokens, $4.13 per 1M output tokens.
How good is Gemini 3.8 Flash?
Its index is 67.8 out of 100. That is place 30 of 283 ranked models.
What is Gemini 3.8 Flash best at?
Knowledge: 70.9 out of 100, place 17 of 268.
How fast is Gemini 3.8 Flash?
68 output tokens per second, 3.0 s to the first token.
What is the context window of Gemini 3.8 Flash?
1.05M tokens.
Is Gemini 3.8 Flash open source?
No. Its weights are not published.
When was Gemini 3.8 Flash released?
September 2026.

Badge

Gemini 3.8 Flash on Inference360: place 30 of 283, index 67.8
HTML
<a href="https://inference360.ai/models/gemini-3-8-flash"><img src="https://inference360.ai/badges/gemini-3-8-flash.svg" alt="Gemini 3.8 Flash on Inference360: place 30 of 283, index 67.8" height="20"></a>
Markdown
[![Gemini 3.8 Flash on Inference360: place 30 of 283, index 67.8](https://inference360.ai/badges/gemini-3-8-flash.svg)](https://inference360.ai/models/gemini-3-8-flash)

Compare with

Grok 4.667.8Claude Opus 4.867.7Step 5 Preview68.4Gemini 3.7 Flash68.5Ling 3.1 Flash67.2

Lower price

Ling 3.1 Flash67.2 · $0.135/1MDeepSeek V4.1 Flash 101768.5 · $0.40/1MClaude Haiku 5.566.3 · $0.50/1M

More from Google

All models
Gemini 3.8 Flash CyberUnrankedGemini 3.8 LiveUnrankedGemini 3.8 Live Extended ThinkingUnrankedGemini 3.7 Flash68.5Gemini 4 Argon78.5