Leading AI in
Predictive, Generative, and Agentic AI.

Generative AI · Image decision models
#1

ImageJevBench

H2O-Lightning-31B

Image decision models answer typed questions about a picture and must return calibrated answers, scored on intelligence, calibration, speed and cost. An open-weights model.

Capability score
80.88
Latency incl. network
0.59s
Per 1,000 decisions
$0.075
H2O vs Major AI Labs
CapabilityHigher is better
  1. #1
    H2O.aiH2O-Lightning-31B
    80.88
  2. #5
    TypeSafe AIJev 1.13.0 + images
    72.44
  3. #12
    OpenAIGPT-6 Luna
    62.90
  4. #20
    CloudflareClef
    58.10
  5. #22
    QwenVisual-Jev (Qwen3.5-4B)
    56.70
  6. #33
    Liquid AId1-3B
    47.80
Go to the ImageJevBench section
Predictive AI · Time-series forecasting
#2

fev-bench

TimeH2O-v1

Zero-shot forecasting on real time series across many domains and frequencies, scored by probabilistic accuracy against a seasonal-naive baseline. 108.8M parameters.

Win rate
89.60%
Skill score
50.80%
Parameters
108.8M
Win rate · top 10
Higher is better
  1. #1
    Continuous Intelligenceci-proto
    90.00%
  2. #2
    H2O.aiTimeH2O-v1
    89.60%
  3. #3
    Mistral AITimestral-209M
    87.10%
  4. #4
    The Forecasting Companyt0-gamma
    84.70%
  5. #5
    Abel AI LabSimons-297M
    82.00%
  6. #6
    GoogleTimesFM-3
    79.30%
  7. #7
    Walmart LabsiAmTime-base-forecast
    73.70%
  8. #8
    AWSChronos-2
    72.20%
  9. #9
    The Forecasting Companyt0-beta
    69.90%
  10. #10
    DatadogToto-2.0-2.5B
    69.40%
Generative AI · Image decision models
#1

ImageJevBench

H2O-Lightning-31B

Image decision models answer typed questions about a picture and must return calibrated answers, scored on intelligence, calibration, speed and cost. An open-weights model.

Capability score
80.88
Latency incl. network
0.59s
Per 1,000 decisions
$0.075
Capability · top 10
Higher is better
Sort by
  1. #1
    H2O.aiH2O-Lightning-31B
    80.88
  2. #2
    Sqwishkettei-fat
    75.89
  3. #3
    Levanto LabsSage 1.4 + images, beta
    75.79
  4. #4
    Nace AIDrex Max
    72.47
  5. #5
    TypeSafe AIJev 1.13.0 + images, preview
    72.44
  6. #6
    AutoloopsAutoloops – Gemma 4 31B IT
    68.35
  7. #7
    Mohit GargImajev-4B
    68.01
  8. #8
    Mohit Gargimajev 9B
    67.07
  9. #9
    Sqwishkettei-chub
    66.50
  10. #10
    H2O.aiH2O-Lightning-4B v1.4.1
    66.16
Generative AI · Image boxes
#3

ImageJevBench Box Track

H2O-Lightning-4B v1.4.1

Models answer with a bounding box and a confidence. Ranks are among 20 ranked systems, scored on 300 public items (546 sealed items are held back for official runs). A preview track: it does not change ImageJevBench's own ranking.

Capability
84.90
Median latency
0.28s
Per 1,000 boxes
$0.096
Capability · top 10
Higher is better
  1. #1
    Z.aiGLM-4.6V
    89.47
  2. #2
    QwenQwen3-VL-32B Instruct
    85.27
  3. #3
    H2O.aiH2O-Lightning-4B v1.4.1
    84.90
  4. #4
    AnthropicClaude Haiku 5.5
    78.81
  5. #5
    GoogleGemma 4 31B IT
    76.44
  6. #6
    QwenQwen3-VL-235B-A22B Instruct
    75.99
  7. #7
    QwenQwen3-VL-30B-A3B Instruct
    75.86
  8. #8
    TypeSafe AIJev 1.14.0
    75.71
  9. #9
    QwenQwen3.5-397B-A17B
    74.69
  10. #10
    QwenQwen3-VL-8B Instruct
    72.78
Generative AI · Decision models
#7

JevBench

H2O-Lightning-31B

Decision models get a situation and typed questions and must return calibrated answers, scored on intelligence, calibration, speed and cost. A 31B model with open weights.

Capability score
84.34
Latency incl. network
0.28s
Per 1,000 decisions
$0.029
Capability · top 10
Higher is better
Sort by
  1. #1
    Torchcast AITorchcast Decision 31B v1.0
    89.45
  2. #2
    Rain WongMetaDecision-31B-Rain
    88.28
  3. #3
    MachinePulse AIKarpo-Decision-31B
    87.16
  4. #4
    Base Labsmammoth-1
    85.23
  5. #5
    PelaAIKnowLine 31B-Gen1
    85.21
  6. #6
    FastinoFastino GLiDE 1.2
    84.92
  7. #7
    H2O.aiH2O-Lightning-31B v1.0.7
    84.34
  8. #8
    AgentScopePiccolo 27B
    83.59
  9. #9
    PerplexityPerplexity Decider v1.1 27B
    82.71
  10. #10
    vLLM Semantic RoutervLLM-SR d3-turbo
    82.59
Predictive AI · Tabular foundation models
#5

TALENT

TabH2O v1.1

Two public benchmark suites of tabular classification and regression datasets, comparing tabular foundation models, AutoML, and tuned gradient boosting across 300 datasets on TALENT and 51 datasets on TabArena.

TALENT rank
#5 / 13
TALENT datasets
300
TabArena rank
#7 / 39
TabArena datasets
51
Try it now

Tabular predictions, one API call away

Open the Playground(opens in a new tab)
TALENT · average rank across datasets · top 10
Lower is better
Benchmark suite
  1. #1
    InriaTabICL v2fine-tuned
    3.48
  2. #2
    LG AI ResearchEXAONE-Tabular
    3.85
  3. #3
    Prior LabsTabPFN v3.0
    4.23
  4. #4
    InriaTabICL v2
    4.40
  5. #5
    H2O.aiTabH2O v1.1
    5.28
  6. #6
    Prior LabsTabPFN v2.6
    5.88
  7. #7
    Prior LabsTabPFN v2.5
    6.23
  8. #8
    AWSAutoGluon
    8.07
  9. #9
    YandexCatBoosttuned
    8.27
  10. #10
    AWSAutoGluonquick preset
    8.46
Agentic AI · Future prediction
#3

FutureX

H2O AI Super Agent

Agents predict real-world events before they resolve, with a new board every week, scored once the answers are known. H2O AI Super Agent has earned 27 FutureX medals (9 first places) across 46 weekly boards entered.

Latest weekly board
#3
FutureX medals
27
First places
9
Boards with a medal
20 / 46
Overall score, weekly board · top 10
Higher is better
  1. #1
    Indiv. AI ResearcherOPS-Agentic-Search_S10.2Kimi K3
    68.06
  2. #2
    Indiv. AI ResearcherSenseFlowDeepSeek-V4-Pro
    66.65
  3. #3
    H2O.aiH2O AI Super Agent v1.82sonnet-5.5
    65.62
  4. #4
    Indiv. AI ResearcherRavenobsidian
    64.83
  5. #5
    Indiv. AI ResearcherOPS-Agentic-Search_S10.7Kimi K3
    64.43
  6. #6
    Indiv. AI ResearcherOPS-Agentic-Search_S10.3Kimi K3
    63.00
  7. #7
    Indiv. AI ResearcherSenseFlowdeepseek-v4-pro-v2
    62.76
  8. #8
    Indiv. AI ResearcherOPS-Agentic-Search_S10.5Kimi K3
    62.44
  9. #9
    Indiv. AI ResearcherNevar-B harnessGarnet
    62.02
  10. #10
    Indiv. AI ResearcherOPS-Agentic-Search_S7Kimi K3
    61.92
Agentic AI · Social forecasting
#2

Social Simulation Arena

H2O AI Super Agent

#3 H2O AI Marina

Entrants forecast the next public release of poll, sentiment and attention series. Forecasts are locked before the answer exists and scored against the published number.

Pairwise win rate
68.99
Questions answered
67
Ahead of Anthropic
+24.72
Pairwise win rate · top 10
Higher is better
  1. #1
    ApodexApodex-FutureFlow
    69.33
  2. #2
    H2O.aiH2O AI Super Agent
    68.99
  3. #3
    H2O.aiH2O AI Marina
    65.04
  4. #4
    AstraculumVac
    54.88
  5. #5
    Social AtomsCrowd (baseline)
    54.52
  6. #6
    AstraculumworldVac
    53.44
  7. #7
    xAIGrok 4.5 · web
    46.09
  8. #8
    GoogleGemini 3.1 Pro · web
    46.05
  9. #9
    AnthropicClaude Sonnet 5 · web+sfc
    44.27
  10. #10
    Renmin University of ChinaYuLan-OneSim
    43.62

Benchmark data read on .