BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases8

Gemini 3 Flash vs Gemini 4 Carbon

Full profiles: Gemini 3 Flash (Google) and Gemini 4 Carbon (Google): every benchmark, price and rank.

Updated · Data from 5 sources

As of 29 Oct 2026, Gemini 3 Flash and Gemini 4 Carbon are neck and neck on quality (59.0 vs 65.1 on the BenchLeader Index, inside the ±9.7 margin), and Gemini 3 Flash is 7.1× cheaper at $1.13 vs $8.00 per 1M tokens blended. They are level in every category we measure both in. Gemini 3 Flash streams ~2× faster.

Key facts (as of 29 Oct 2026)

  • Gemini 3 Flash scores 59.0 on the BenchLeader Index and Gemini 4 Carbon scores 65.1 (BenchLeader, 29 Oct 2026).
  • Gemini 3 Flash costs $1.13 per 1M tokens blended (list price, 29 Oct 2026).
  • Gemini 4 Carbon costs $8.00 per 1M tokens blended (list price, 29 Oct 2026).
  • Gemini 3 Flash streams ~2× faster than Gemini 4 Carbon (Artificial Analysis, 7-day median, 29 Oct 2026).
  • Of the 4 benchmarks both models have, Gemini 4 Carbon leads 4 (BenchLeader, 29 Oct 2026).
Gemini 3 Flash benchmarks thinking Google Gemini 4 Carbon benchmarks high GoogleOpen in the comparison tool

Which to pick for…

The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.

Use casePickWhyConfidence
Long documentsEither64 vs 65 category score; AA-LCR v1.1 78.0% vs 79.3%Either
Lowest costGemini 3 Flash$1.13 vs $8.00 per 1M blended; agentic coding session $0.042 vs $0.320–
Fastest first answerGemini 3 Flash6.0 s (Gemini 3 Flash (thinking)) vs 103.1 s (Gemini 4 Carbon (high)) to the first answer token–
High volume on a budgetNeither flagshipA smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs–

Key differences

  • Gemini 3 Flash (thinking) and Gemini 4 Carbon (high) are neck and neck on quality (59.0 vs 65.1, within the ±9.7 margin).
  • Gemini 3 Flash (thinking) is stronger in agents & tools, instruction following, multimodal.
  • Gemini 4 Carbon (high) is stronger in long context, coding, knowledge, reasoning.
  • Gemini 3 Flash (thinking) is 7.1× cheaper ($1.13 vs $8.00 per 1M blended).
  • Gemini 3 Flash (thinking) streams 2.2× faster (215 vs 100 tokens per second).

At a glance

SpecGemini 3 FlashGemini 4 Carbon
DeveloperGoogleGoogle
Released17 Dec 202528 Oct 2026
Context window (tokens)1,048,5761,000,000
Compared atGemini 3 Flash (thinking)Gemini 4 Carbon (high)
Open weightsNoNo
Price in / out / cached input per 1M$0.500 / $3.00 / $0.050$4.00 / $20.00 / –

Category by category

Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.

CategoryGemini 3 Flash (thinking)Gemini 4 Carbon (high)Leader
Reasoning–78.9not comparable
Coding–71.7not comparable
Agents & tools55.1–not comparable
Knowledge–75.8not comparable
Instruction following75.8–not comparable
Multimodal62.7–not comparable
Long context63.864.5level

Where sources disagree

Across the 4 benchmarks both have, Gemini 3 Flash leads 0, Gemini 4 Carbon leads 4.

Overall indexGemini 3 Flash (thinking)Gemini 4 Carbon (high)Leader
BenchLeader Index59.065.1neck and neck

Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).

Price, speed and latency

List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.

MeasureGemini 3 Flash (thinking)Gemini 4 Carbon (high)
Blended price per 1M tokens$1.13$8.00
Input / output per 1M$0.500 / $3.00$4.00 / $20.00
Cached input per 1M$0.050–
Price per task: chat reply$0.0011$0.0076
Price per task: summarize a 30-page report$0.0078$0.060
Price per task: code edit$0.0075$0.054
Price per task: agentic coding session$0.042$0.320
Price per task: structured extraction$0.0016$0.012
Cost to run a full benchmark suite–$7.76
Output speed215 tok/s100 tok/s
Time to first answer6.0 s103.1 s
Time to a complete answer8.3 s108.1 s

Each model is compared at its best-ranked reasoning-effort setting. Gemini 3 Flash (thinking) scores 59.0 and takes 6 s to its first answer token; Gemini 3 Flash (high) scores 58.3 in 1 s, Gemini 3 Flash scores 57.0 in 1 s. Gemini 4 Carbon (high) scores 65.1 and takes 103 s to its first answer token. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.

Every benchmark

114 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.

MetricPublisherGemini 3 Flash (thinking)Gemini 4 Carbon (high)
BenchLeader IndexBenchLeader59.065.1
Agents & tools scoreBenchLeader55.1–
Instruction following scoreBenchLeader75.8–
Long context scoreBenchLeader63.864.5
Multimodal scoreBenchLeader62.7–
Coding scoreBenchLeader–71.7
Knowledge scoreBenchLeader–75.8
Reasoning scoreBenchLeader–78.9
Blended price $/MBenchLeader$1.13$8.00
Output speedBenchLeader215 tok/s100 tok/s
Time to first answerBenchLeader6.0 s103.1 s
Context windowBenchLeader1,048,576 tokens1,000,000 tokens
Reasoning
GPQA DiamondEpoch AI Benchmarking Hub83.2%†–
SimpleBenchSimpleBench61.1%†–
Arena Hard PromptsArena (LMArena)1493†–
GPQA Diamond (AA)Artificial Analysis89.8%–
Humanity's Last Exam (AA)Artificial Analysis36.6%60.4%
HLE-Diamond (AA)Artificial Analysis–69.1%
GPQA Diamond (Vals)Vals AI87.9%†–
ARC-AGI-1ARC Prize84.7%†–
ARC-AGI-2ARC Prize33.6%†–
CritPtArtificial Analysis8.6%36.0%
CritPt v2 hard subset (AA)Artificial Analysis–63.0%
SciPredictScale AI SEAL22.2%†–
Chess PuzzlesEpoch AI Benchmarking Hub39.0%†–
Mystery Game PuzzlesEpoch AI Benchmarking Hub20.0%†–
LMCAEpoch AI Benchmarking Hub43.1%†–
DTBenchEpoch AI Benchmarking Hub89.1%†–
ForecastBenchForecastBench64.7%†–
Coding
SWE-bench Verified (Epoch)Epoch AI Benchmarking Hub75.4%†–
SciCodeSciCode–64.2%
WeirdMLWeirdML61.6%†–
GSO-BenchGSO-Bench9.8%†–
Arena CodingArena (LMArena)1506†–
Arena WebDevArena (LMArena)1440†–
LiveCodeBenchVals AI85.6%†–
SWE-bench (Vals)Vals AI75.0%†–
Show all 114 benchmarks
MetricPublisherGemini 3 Flash (thinking)Gemini 4 Carbon (high)
Coding
SWE-Bench ProScale AI SEAL34.6%†–
SWE-bench Verified (bash only)SWE-bench75.8%†–
SWE-bench Verified (any scaffold)SWE-bench75.8%†–
Code MigrationVals AI6.4%†–
ProgramBenchVals AI0.0%†–
Vibe Code Bench v1.1Vals AI20.2%†89.1%
SWE-Bench Pro (private)Scale AI SEAL34.6%†–
SWE Atlas: Codebase QnAScale AI SEAL8.2%†–
SWE Atlas: RefactoringScale AI SEAL10.0%†–
SWE Atlas: Test WritingScale AI SEAL30.3%†–
ALE-BenchEpoch AI Benchmarking Hub1367.2†–
Terminal-Bench 4.0 (AA)Artificial Analysis–65.2%
FrontierCode 1.1 (AA)Artificial Analysis–48.5%
FrontierCode 2.0 (AA)Artificial Analysis–40.6%
Agents & tools
Terminal-BenchTerminal-Bench64.3%†–
Terminal-Bench HardArtificial Analysis38.6%–
τ²-Bench Telecom (AA)Artificial Analysis80.4%–
Terminal-Bench 2.1 (Vals)Vals AI53.9%†–
MCP AtlasScale AI SEAL62.0%†–
τ²-benchτ²-bench83.5%†–
τ³-bench Bankingτ²-bench27.3%†–
τ³-Banking (AA)Artificial Analysis20.8%–
APEX-Agents (AA)Artificial Analysis27.7%–
Finance Agent v2Vals AI42.5%†–
Finance Agent v2.1Vals AI–67.2%
Harvey's Legal Agent Benchmark v1.1Vals AI–17.5%
Harvey's Legal Agent Benchmark v1Vals AI0.0%†–
Legal Research BenchVals AI18.3%†55.8%
Poker AgentVals AI1100.2%†–
Tax Agent BenchVals AI11.8%†48.9%
Terminal-Bench 2.0 (Vals)Vals AI51.7%†–
Terminal-Bench 4.0 (Vals)Vals AI–63.1%
BALROGBALROG48.1%†–
DeepResearch BenchDeepResearch Bench47.9%†–
Vending-Bench 2Andon Labs3634.7†16484.2†
AutomationBenchArtificial Analysis–77.8%
GDP.pdf (AA)Artificial Analysis–28.4%
AA-Briefcase v1.1Artificial Analysis–1609
Terminal-Bench-Science 0.1 (AA)Artificial Analysis–64.8%
BrowseCompOpenAI16.9%†–
FACTS SearchGoogle80.9%†–
Maths
FrontierMath Tiers 1–3Epoch AI Benchmarking Hub51.2%†–
FrontierMath Tier 4Epoch AI Benchmarking Hub17.1%†–
OTIS Mock AIMEEpoch AI Benchmarking Hub92.8%†–
ProofBenchVals AI15.0%†–
AIME (Vals)Vals AI95.6%†–
MGSMVals AI93.3%†–
AIME 2026MathArena96.7%†–
HMMT February 2026MathArena89.4%†–
MathArena ApexMathArena15.6%†–
Arena MathsArena (LMArena)1475†–
Knowledge
SimpleQA VerifiedEpoch AI Benchmarking Hub66.8%†–
AA-Omniscience v1.1Artificial Analysis–53.5
MMLU-ProVals AI88.6%†–
LegalBenchVals AI86.9%†–
CorpFinVals AI66.4%†–
TaxEvalVals AI73.9%†–
MedQAVals AI95.8%†–
CaseLaw v2Vals AI55.8%†–
Excel Modeling BenchmarkVals AI26.0%†–
Excel Modeling Benchmark v1.1Vals AI–76.5%
MedCodeVals AI55.9%†–
MedScribeVals AI69.9%†–
GDP.pdfEpoch AI Benchmarking Hub10.0%†–
AA-Omniscience v1.1: accuracyArtificial Analysis–58.3%
AA-Omniscience v1.1: non-hallucinationArtificial Analysis–88.6%
Global-MMLU-Lite (AA)Artificial Analysis92.8%†–
Instruction following
IFBenchArtificial Analysis78.0%–
Arena Instruction FollowingArena (LMArena)1458†–
Human preference
Arena TextArena (LMArena)1473†–
Arena Creative WritingArena (LMArena)1457†–
Arena Multi-turnArena (LMArena)1482†–
Arena Longer QueriesArena (LMArena)1476†–
Multimodal
Arena VisionArena (LMArena)1285†–
MMMU-ProArtificial Analysis79.9%–
MMMU-Pro (Vals)Vals AI87.6%†–
MortgageTaxVals AI68.7%†–
GeoBenchGeoBench88.0%†–
VPCTEpoch AI Benchmarking Hub72.6%†–
Blueprint-Bench 2Epoch AI Benchmarking Hub0.0%†58.6%†
Long context
AA-LCR v1.1Artificial Analysis78.0%79.3%
Arena DocumentArena (LMArena)1439†–
Composite
Epoch Capabilities IndexEpoch AI Benchmarking Hub151.9†–
AA Intelligence Index v5.0Artificial Analysis–59.4
AA Intelligence Index v4.3.2Artificial Analysis26.3–
AA Progress Index v1Artificial Analysis290.1351.6
Vals IndexVals AI–70.4
Vals Finance IndexVals AI–66.3%
Vals Multimodal IndexVals AI52.2%†–

Data as of 2026-10-29. Best configuration of each model; every score links to its source on the model pages.

What changed

  • Gemini 3 Flash (thinking): BenchLeader Index 60.1 → 59.0.
  • Gemini 3 Flash (thinking): 1 new result (AA Progress Index v1).
  • Gemini 3 Flash (thinking): 2 new results (AA-Omniscience v1.1: accuracy, AA-Omniscience v1.1: non-hallucination).
  • Gemini 3 Flash (thinking): 2 new results (MLCR, Global-MMLU-Lite (AA)).
  • Gemini 3 Flash (thinking): BenchLeader Index 60.7 → 60.1.

Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs

Gemini 3 Flash vs Gemini 4 Carbon: questions

Is Gemini 3 Flash better than Gemini 4 Carbon?
Gemini 3 Flash (thinking) and Gemini 4 Carbon (high) are neck and neck on quality (59.0 vs 65.1, within the ±9.7 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 29 Oct 2026, so check the category scores for your use.
Gemini 3 Flash vs Gemini 4 Carbon: which is cheaper, and will that change?
Gemini 3 Flash is cheaper today: $1.13 against $8.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
Which is faster, Gemini 3 Flash or Gemini 4 Carbon?
Gemini 3 Flash streams ~2× faster (215 vs 100 output tokens per second, Artificial Analysis 7-day median).
Which is better for long documents?
They are level on long-context benchmarks (64 vs 65). Both take about 1M tokens: Gemini 3 Flash 1,048,576, Gemini 4 Carbon 1,000,000.
Which has lower latency?
Gemini 3 Flash answers first: 6.0 s against 103.1 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
More questions (2)
Do independent benchmarks agree on Gemini 3 Flash vs Gemini 4 Carbon?
Across the 4 benchmarks both have, Gemini 3 Flash leads 0, Gemini 4 Carbon leads 4.
When was Gemini 4 Carbon released?
Gemini 4 Carbon was released on 28 Oct 2026.