priorsBuilder

Google · released 2026-09-02

Gemini 3.8 Flash

Measured on 102 benchmarks across 3 reasoning efforts.

General frontier

46.8

#17 of 32 models · best at high

Reasoning efforts in your mix

high46.8#58 of 129
medium45.2#66 of 129
low40.6#83 of 129

Your mix’s benchmarks

7 of 8 measured

reasoning

Humanity's Last Exam

587 results
  1. #36high of 58747.8%
  2. #71medium of 58742.1%
  3. #104low of 58737.1%

Leader: Opus 5.5 · 61.4%

knowledge

AA-Omniscience

526 results
  1. #34high of 52629.6
  2. #37medium of 52628.6
  3. #59low of 52621.4

Leader: Opus 5.5 · 46.4

reasoning

ARC-AGI-3

61 results
  1. #12high of 6135.0%
  2. #14medium of 6124.2%
  3. #16low of 6115.1%

Leader: GPT-6 Astra · 99.9%

coding

DeepSWE v1.1

70 results
  1. #2high of 7073.8%
  2. #10medium of 7071.0%

Leader: GPT-6 Astra · 74.1%

math

FrontierMath Tier 4 (v2)

70 results
  1. #46high of 7022.0%

Leader: GPT-6.1 Sol · 100.0%

reasoning

SimpleBench

97 results
  1. #6max of 9782.4%

Leader: Opus 5.5 · 88.4%

agentic

Terminal-Bench 4.0

35 results
  1. #29high of 3519.1%

Leader: Opus 5.5 · 64.8%

All 102 benchmarks

show 95 more

agentic

AA-Briefcase v1.1

224 results
  1. #86high of 2241203
  2. #90medium of 2241181
  3. #109low of 2241062

Leader: Sonnet 5.5 · 1823

agentic

Agent Arena

50 results
  1. #23high of 502.8%

Leader: Opus 5.5 · 15.5%

agentic

AutomationBench (Zapier)

115 results
  1. #17medium of 11529.7%
  2. #18high of 11529.7%

Leader: Gemini 4 · 51.3%

agentic

AutomationBench-AA

219 results
  1. #30medium of 21960.9%
  2. #34high of 21959.9%
  3. #99low of 21936.5%

Leader: Gemini 4 · 77.5%

agentic

BioMysteryBench

27 results
  1. #23high of 2762.2%

Leader: Sonnet 5.5 · 81.1%

agentic

CUA-bench

8 results
  1. #8high of 84.2%

Leader: GPT-6 Astra · 19.2%

agentic

Harvey's Legal Agent Benchmark

79 results
  1. #19high of 7910.0%

Leader: Muse Spark 1.2 · 25.4%

agentic

ITBench-AA

45 results
  1. #3high of 4552.5%

Leader: GPT-5.6 Sol · 56.2%

agentic

Legal Research Bench

78 results
  1. #32high of 7838.9%

Leader: Muse Spark 1.3 · 55.3%

agentic

SkillsBench

34 results
  1. #11high of 3458.0%

Leader: DeepSeek V4.1 Flash · 69.8%

agentic

Tax Agent Bench

69 results
  1. #18high of 6932.1%

Leader: Fable 5.1 · 49.2%

agentic

Terminal-Bench 4.0 (Vals)

47 results
  1. #26high of 4719.7%

Leader: Opus 5.5 · 65.2%

agentic

Terminal-Bench Science (Vals)

40 results
  1. #18high of 408.6%

Leader: GPT-6 Astra · 62.9%

agentic

Time Horizon Index: KSP

13 results
  1. #6high of 1318.8%

Leader: Opus 5.5 · 91.3%

agentic

Vending-Bench 2

67 results
  1. #30max of 67$5,094

Leader: GPT-6 Astra · $15,515

agentic

WeirdML v3

17 results
  1. #12high of 177.4%

Leader: GPT-6 Astra · 42.2%

agentic

τ³-Banking

206 results
  1. #13medium of 20645.8%
  2. #15high of 20644.9%
  3. #55low of 20633.2%

Leader: Qwen 3.8 Max · 51.3%

coding

Arena Image-to-WebDev

58 results
  1. #18high of 581593

Leader: Opus 5.5 · 1749

coding

Arena WebDev

122 results
  1. #30high of 1221583

Leader: Opus 5.5 · 1813

coding

Code Migration

77 results
  1. #35high of 7736.5%

Leader: Sonnet 5.5 · 69.8%

coding

FrontierCode 1.1 Extended

132 results
  1. #75medium of 13253.4%
  2. #85high of 13251.2%

Leader: Opus 5.5 · 65.3%

coding

FrontierCode 1.1 Main

132 results
  1. #67medium of 13241.2%
  2. #79high of 13238.0%

Leader: Opus 5.5 · 54.6%

coding

FrontierSWE v2

21 results
  1. #16max of 2119.6%

Leader: GPT-6 Astra · 65.5%

coding

IOI

44 results
  1. #19high of 4456.9%

Leader: Gemini 4 · 100.0%

coding

LiveCodeBench (Vals)

127 results
  1. #3high of 12789.5%

Leader: Fable 5.1 · 90.5%

coding

ProgramBench v1 Almost Resolved

62 results
  1. #21high of 6214.0%

Leader: Opus 5.5 · 65.0%

coding

ProgramBench v1 Fully Resolved

62 results
  1. #15high of 621.0%

Leader: Opus 5.5 · 18.5%

coding

ProgramBench v1 Raw Pass Rate

62 results
  1. #17high of 6271.9%

Leader: Opus 5.5 · 87.0%

coding

SciCode

226 results
  1. #34high of 22656.6%
  2. #51medium of 22655.1%
  3. #55low of 22655.0%

Leader: Opus 5.5 · 66.9%

coding

SWE Atlas-QnA

22 results
  1. #6max of 2247.0%

Leader: Opus 5 · 63.2%

coding

SWE-bench Verified (Vals)

85 results
  1. #23high of 8580.0%

Leader: Opus 5 · 97.0%

coding

Terminal-Bench 2.1 (Vals)

74 results
  1. #8high of 7481.3%

Leader: Opus 5.5 · 87.6%

coding

Terminal-Bench Science-AA

47 results
  1. #24high of 4710.0%

Leader: GPT-6 Astra · 63.3%

coding

Terminal-Bench v2.1-AA

238 results
  1. #19high of 23887.6%
  2. #35medium of 23883.9%
  3. #37low of 23883.1%

Leader: Fable 5.1 · 91.4%

coding

Terminal-Bench v4-AA

216 results
  1. #61high of 21619.7%
  2. #62medium of 21619.7%
  3. #87low of 21610.1%

Leader: Sonnet 5.5 · 63.6%

coding

Vibe Code Bench 1-100

22 results
  1. #10high of 2218.8%

Leader: Opus 5.5 · 30.4%

coding

Vibe Code Bench v1.1

105 results
  1. #27high of 10578.7%

Leader: Sonnet 5.5 · 92.4%

coding

WeirdML v2

127 results
  1. #15max of 12784.8%

Leader: GPT-6 Astra Pro · 93.6%

composite

AA Intelligence Index v4.3.2

610 results
  1. #57high of 61040.9
  2. #63medium of 61039.8
  3. #103low of 61033.5

Leader: Opus 5.5 · 57.6

composite

Arena Text

220 results
  1. #8high of 2201497

Leader: Gemini 4 · 1525

composite

Arena Text — Multi-Turn

220 results
  1. #8high of 2201503

Leader: Gemini 4 · 1553

composite

Vals Index

47 results
  1. #17high of 4754.9%

Leader: Gemini 4 · 68.9%

composite

Vals RSI Index v1.1

24 results
  1. #16max of 2419.9%

Leader: Opus 5.5 · 37.3%

composite

Vals RSI Index v1.1: Harness Engineering: Judge

24 results
  1. #13max of 2414.5%

Leader: GPT-6 Sol · 22.8%

composite

Vals RSI Index v1.1: Post-training: Finance Agent

24 results
  1. #18max of 240.0%

Leader: Opus 5.5 · 35.1%

composite

Vals RSI Index v1.1: Pre-training: Compression

24 results
  1. #17max of 2429.0%

Leader: Opus 5.5 · 46.5%

composite

Vals RSI Index v1.1: Pre-training: LM Training

24 results
  1. #17max of 2436.2%

Leader: Opus 5.5 · 53.0%

cost

AA Intelligence Index Cost per Task

177 results
  1. #111medium of 177$0.93
  2. #127high of 177$1.24

Leader: GPT-6 Luna · $0.00

cost

Cost to Run AA Intelligence Index

177 results
  1. #101medium of 177$1,100
  2. #126high of 177$1,623

Leader: GPT-6 Luna · $10.63

creative

MineBench

72 results
  1. #14max of 721842

Leader: GPT-6 Astra Pro · 2281

creative

threejseval

34 results
  1. #13high of 341580

Leader: Opus 5.5 · 2079

creative

VoxelBench (text)

55 results
  1. #20max of 551786

Leader: GPT-6 Astra · 2733

economics

Excel Modeling Benchmark

74 results
  1. #8high of 7472.2%

Leader: Fable 5.1 · 76.7%

economics

Finance Agent v2

78 results
  1. #2high of 7861.4%

Leader: Gemini 4 · 65.4%

economics

GDP.pdf

210 results
  1. #46medium of 21022.8%
  2. #56high of 21021.0%
  3. #71low of 21018.6%

Leader: Muse Wonder Full (Max) · 32.6%

economics

GDPval-AA v2.1

288 results
  1. #63high of 2881436
  2. #67medium of 2881422
  3. #97low of 2881307

Leader: Opus 5.5 · 1866

economics

MortgageTax (Vals)

84 results
  1. #41high of 8465.3%

Leader: Opus 5 · 72.1%

economics

TaxEval (Vals)

125 results
  1. #36high of 12574.4%

Leader: Muse Spark 1.2 · 80.4%

knowledge

AA-Omniscience Accuracy

526 results
  1. #39high of 52654.6%
  2. #49medium of 52653.0%
  3. #51low of 52652.2%

Leader: Fable 5.1 · 67.2%

knowledge

AA-Omniscience Non-hallucination

526 results
  1. #125medium of 52648.1%
  2. #139high of 52644.8%
  3. #181low of 52635.4%

Leader: MiniCPM5-1B (Non-reasoning) · 99.1%

knowledge

GPQA Diamond

539 results
  1. #3high of 53995.3%
  2. #19medium of 53993.5%
  3. #43low of 53992.0%

Leader: GPT-6 Astra · 96.3%

knowledge

GPQA Diamond (Vals)

122 results
  1. #4high of 12294.4%

Leader: Gemini 3.1 Pro · 95.5%

knowledge

Harvey LAB-AA v1.1

27 results
  1. #22high of 270.0%

Leader: Muse Wonder Full (Max) · 13.1%

knowledge

MedCode

94 results
  1. #29high of 9448.1%

Leader: Opus 5 · 63.6%

knowledge

MedScribe

96 results
  1. #33high of 9684.5%

Leader: Opus 5.5 · 91.4%

knowledge

MMLU Pro (Vals)

122 results
  1. #5high of 12290.2%

Leader: Fable 5.1 · 92.4%

knowledge

MMMU Pro (Vals)

81 results
  1. #4high of 8189.1%

Leader: Fable 5.1 · 90.6%

knowledge

MMMU-Pro

265 results
  1. #9high of 26585.6%
  2. #19low of 26584.5%
  3. #21medium of 26584.2%

Leader: Opus 5.5 · 87.7%

knowledge

Public Benefits Bench v1.1

50 results
  1. #22high of 5065.3%

Leader: Opus 5 · 76.9%

long-context

AA-LCR v1.1

523 results
  1. #13medium of 52384.0%
  2. #54high of 52381.3%
  3. #66low of 52380.7%

Leader: Kimi K3 · 88.7%

long-context

MLCR-AA

94 results
  1. #25high of 9421.7%

Leader: Sonnet 5.5 · 75.0%

math

FrontierMath Tiers 1–3 (v2)

115 results
  1. #26high of 11568.4%

Leader: GPT-6.1 Sol · 93.7%

math

ProofBench v1.1

52 results
  1. #34high of 5248.0%

Leader: Sonnet 5.5 · 100.0%

other

Arena Vision

116 results
  1. #13high of 1161289

Leader: Fable 5 · 1308

other

T3 Code usage share

85 results
  1. #19max of 850.0%

Leader: Opus 5.5 · 29.1%

reasoning

ARC-AGI-1

215 results
  1. #6high of 21598.5%
  2. #19medium of 21597.5%
  3. #68low of 21590.5%

Leader: Fable 5 · 98.5%

reasoning

ARC-AGI-2

216 results
  1. #18high of 21689.2%
  2. #36medium of 21682.9%
  3. #40low of 21677.5%

Leader: GPT-6 Astra · 95.0%

reasoning

Blueprint-Bench 2

28 results
  1. #7max of 2838.6%

Leader: Gemini 4 · 54.4%

reasoning

CritPt

529 results
  1. #67high of 52918.3%
  2. #102medium of 52912.3%
  3. #163low of 5294.0%

Leader: Muse Wonder Full (Max) · 32.9%

reasoning

Furniture Assembly

31 results
  1. #22high of 3131.7%

Leader: Opus 5.5 · 83.3%

reasoning

Humanity's Sixth Sense

25 results
  1. #4high of 2541.6%

Leader: GPT-6 Astra · 53.6%

reasoning

LegalBench

135 results
  1. #7high of 13587.0%

Leader: Fable 5 · 88.6%

reasoning

MysteryMechanism

27 results
  1. #8high of 2736.5%

Leader: GPT-6 Astra · 53.2%

reasoning

Roboflow Visual Reasoning

102 results
  1. #5high of 10284.5%
  2. #11low of 10281.2%

Leader: GPT-6 Astra · 91.2%

reasoning

SAGE

80 results
  1. #65high of 8035.1%

Leader: Opus 4.7 · 56.1%

reasoning

SimpleQA Verified

47 results
  1. #3max of 4774.6%

Leader: Gemini 3.1 Pro · 77.5%

reliability

Tool-call error rate

172 results
  1. #120max of 1722.9%

Leader: Gemini 4 · 0.0%

reliability

Uptime (7d)

182 results
  1. #143max of 18298.3%

Leader: GPT-5 Pro · 100.0%

safety

CWE-Bench

19 results
  1. #13high of 1952.5%

Leader: Grok 4.7 · 68.3%

safety

CyberBench v1.1

45 results
  1. #41high of 4543.8%

Leader: Wonder · 80.4%

safety

SRE Bench

32 results
  1. #11high of 3211.1%

Leader: GPT-6 Astra · 56.9%

speed

Latency (Time to First Token)

351 results
  1. #288high of 35124s

Leader: Gemini 2.5 Flash Lite · 0.3s

speed

Output Speed (tokens/s)

351 results
  1. #98high of 351145.4

Leader: Celeris-1 · 1564.1

speed

Provider latency

178 results
  1. #140max of 1782.7s

Leader: Qwen 2.5 7B · 0.2s

speed

Provider throughput

176 results
  1. #45max of 17685.5

Leader: Muse Spark 1.1 · 204.5