Models and agents you can run in your own environment, measured on the public leaderboards where the field competes.
| Kind | Benchmark | H2O entry | Rank | Link |
|---|---|---|---|---|
| Generative | JevBench Capability | H2O-Lightning-31B | #1 of 159 | Model card ↗ |
| Generative | ImageJevBench Capability | H2O-Lightning-31B | #1 of 54 | Model card ↗ |
| Predictive | fev-bench | TimeH2O-v1 | #1 of 32 | Model card ↗ |
| Agentic | FutureX | H2O AI Super Agent | #1 of 24 | Enterprise h2oGPTe ↗ |
| Agentic | Social Simulation Arena | H2O AI Super Agent | #1 of 24 | Enterprise h2oGPTe ↗ |
| Agentic | Social Simulation Arena | H2O AI Marina | #3 of 24 | H2O AI Marina ↗ |
| Predictive | TALENT | TabH2O | #5 of 13 | Try it live ↗ |
| Predictive | TabArena | TabH2O | #7 of 39 | Try it live ↗ |
JevBench · ImageJevBench · fev-bench · Social Simulation Arena · FutureX · TabH2O
Generative AI · Decision models
H2O-Lightning-31B ranks #1 of 159 on JevBench Capability, the board's headline score
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 | latency incl. network, board estimate | per 1,000 decisions, board estimate |
|---|---|---|
| 84.3 | 0.28 s | $0.044 |
Leaderboard ↗ · How it scores ↗ · H2O blog post ↗ · Model card ↗
Capability, latency and cost, side by side · Capability top 10 · sorted by Capability · higher is better
Capability, latency and cost, side by side
| # | Capability · higher is better | Model | Latency · lower is better | Cost · lower is better |
|---|---|---|---|---|
| 1 | 84.3 | H2O-Lightning-31B v1.0.7 | 280 ms | $0.044 |
| 2 | 82.1 | basal-omni-1.0-ultra | 370 ms | $0.045 |
| 3 | 81.8 | Arcbright Elf 1 31B | 270 ms | $0.047 |
| 4 | 81.7 | Quyet-1.0-Large | 390 ms | $0.045 |
| 5 | 81.5 | Jeff-1.0-Large | 320 ms | $0.049 |
| 6 | 79.6 | metask-jev-rain-31B-trained service v3 | 520 ms | $0.049 |
| 7 | 79.6 | decisio v0.8.0 on gemma-4-31B-it | 240 ms | $0.041 |
| 8 | 78.9 | metask-jev-rain-31B-trained | 620 ms | $0.044 |
| 9 | 77.9 | Bobcat 1.3 | 290 ms | $0.047 |
| 10 | 77.6 | deck-31B | 410 ms | $0.048 |
| 77.1 | Jev 1.13.0 reference · closed API · not ranked |
240 ms | $0.032 | |
| 15 | 75.0 | H2O-Lightning-4B v1.1 | 210 ms | $0.016 |
| 75.0 | Jev 1.14.0 reference · closed API · not ranked |
230 ms | $0.032 |
Hatched: the board's Jev reference, a closed API listed but not ranked. Capability, the board's headline score, ranks the 159 inside its cost and latency caps. On the board's official Composite for open-weights systems H2O-Lightning-31B is #3 of 220. Latency includes network: for self-hosted models the board estimates it as 2 × its measured GPU time + 150 ms. Cost is the board's estimate. Benchmark Heaven, board v1.7.48.
Generative AI · Image decision models
H2O-Lightning-31B ranks #1 of 54 on ImageJevBench Capability, the board's headline score
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 | latency incl. network, board estimate | per 1,000 decisions, at the board's GPU rate |
|---|---|---|
| 80.9 | 0.59 s | $0.075 |
Leaderboard ↗ · How it scores ↗ · Model card ↗
Capability, latency and cost, side by side · Capability top 10 · sorted by Capability · higher is better
Capability, latency and cost, side by side
| # | Capability · higher is better | Model | Latency · lower is better | Cost · lower is better |
|---|---|---|---|---|
| 1 | 80.9 | H2O-Lightning-31B | 590 ms | $0.075 |
| 2 | 75.9 | kettei-fat | 420 ms | $0.011 |
| 3 | 75.8 | Sage 1.4 + images, beta | 560 ms | $0.079 |
| 4 | 72.5 | Drex Max | 850 ms | $0.084 |
| 5 | 72.4 | Jev 1.13.0 + images, preview | 380 ms | $0.16 |
| 6 | 68.3 | Autoloops – Gemma 4 31B IT | 1190 ms | $0.083 |
| 7 | 68.0 | Imajev-4B | 370 ms | $0.038 |
| 8 | 67.1 | imajev 9B | 670 ms | $0.089 |
| 9 | 66.5 | kettei-chub | 310 ms | $0.0073 |
| 10 | 64.8 | Surogate Rune 26B-A4B v3 | 490 ms | $0.058 |
Capability, the board's headline score, ranks the 54 Jev-class systems: those inside 2 × the cost and median latency of the board's Jev reference. On the board's official Composite H2O-Lightning-31B is #5 of 63. Latency includes network: for self-hosted models the board estimates it as 2 × its measured GPU time + 150 ms. A self-hosted system's cost is its measured GPU time at the board's hourly rate. Benchmark Heaven, board v0.3.0.
Predictive AI · Time-series forecasting
TimeH2O-v1 ranks #1 of 32 on the fev-bench leaderboard, by win rate
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 | skill score | parameters |
|---|---|---|
| 91.9% | 50.8% | 108.8M |
Leaderboard ↗ · fev-bench paper ↗ · Model card ↗
Win rate · top 8 of 32 · higher is better
Win rate · top 8 of 32
| Rank | Entry | Win rate |
|---|---|---|
| #1 | TimeH2O-v1 | 91.9% |
| #2 | Timestral-209M | 89.6% |
| #3 | t0-gamma | 88.1% |
| #4 | TimesFM-3 | 82.7% |
| #5 | Chronos-2 | 76.8% |
| #6 | t0-beta | 74.5% |
| #7 | TiRex-2 | 73.5% |
| #8 | Toto-2.0-2.5B | 73.0% |
Default leaderboard view. With the closed-API models included it is #2 of 36. TimeH2O-v1 has a non-commercial licence.
Agentic AI · Social forecasting
H2O AI Super Agent ranks #1 of 24 on Social Simulation Arena's Season 0 pairwise win rate
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 | questions answered | H2O AI Super Agent | H2O AI Marina |
|---|---|---|---|
| 67.29 | 43 | #1 | #3 |
Leaderboard ↗ · How scoring works ↗ · Enterprise h2oGPTe ↗ · H2O AI Marina ↗
Pairwise win rate · top 8 of 24 · higher is better
Pairwise win rate · top 8 of 24
| Rank | Entry | Pairwise win rate |
|---|---|---|
| #1 | H2O AI Super Agent | 67.29 |
| #2 | Apodex-FutureFlow | 66.60 |
| #3 | H2O AI Marina | 62.20 |
| #4 | Vac | 57.08 |
| #5 | Claude Sonnet 5 · web+sfc | 56.89 |
| #6 | Crowd (baseline) | 54.95 |
| #7 | worldVac | 54.11 |
| #8 | Gemini 3.1 Pro · web | 52.12 |
The site's default view: Season 0 pairwise win rate, 43 questions locked from 30 Sep 2026, one row per method. By arena score over every resolved number question Super Agent is #3 of 25 and Marina #4; by arena score on the Season 0 main board Super Agent is #5 of 26 and Marina #7. Entrants answer different numbers of questions.
Agentic AI · Future prediction
H2O AI Super Agent ranks #1 of 24 on FutureX's weekly board for 2026 Oct. Week 2
Agents predict real-world events before they resolve, with a new board every week, scored once the answers are known.
| latest weekly board | first places | weekly boards with a medal, of 43 entered |
|---|---|---|
| #1 of 24 | 8 | 18 |
Leaderboard ↗ · FutureX paper ↗ · Enterprise h2oGPTe ↗
Overall score, weekly board · top 8 of 24 · higher is better
Overall score, weekly board · top 8 of 24
| # | Model | Framework | Organization | Score |
|---|---|---|---|---|
| 1 | sonnet-5.5 - v2 | H2O AI Super Agent v1.82 | h2o.ai | 52.16 |
| 2 | sonnet-5.5 - 3x | H2O AI Super Agent v1.82 | h2o.ai | 46.60 |
| 3 | GLM 5.2 | arce #1 | Indiv. AI Researcher | 46.05 |
| 4 | grok-4.7+sonnet-5.5 - v4 | H2O AI Super Agent v1.82 | h2o.ai | 45.56 |
| 5 | sonnet-5-5 | H2O AI Super Agent v1.82 | h2o.ai | 45.49 |
| 6 | grok-4.7+sonnet-5.5 - v1 | H2O AI Super Agent v1.82 | h2o.ai | 45.49 |
| 7 | GLM 5.2 | arce #2 | Indiv. AI Researcher | 45.00 |
| 8 | blue-fish | blue-fish | Google DeepMind | 43.88 |
Best of 8 H2O submissions. The agent runs on third-party base LLMs; medals count every H2O entry.
Predictive AI · Tabular foundation models
TabH2O ranks #5 of 13 on TALENT and #7 of 39 on TabArena
Two public benchmark suites of tabular classification and regression datasets, comparing tabular foundation models, AutoML and tuned gradient boosting.
| TALENT | TabArena |
|---|---|
| #5 of 13 | #7 of 39 |
TALENT ↗ · TabArena ↗ · Introducing TabH2O ↗ · TabH2O paper ↗ · TabH2O ↗
Try it now
Run TabH2O on your own table in the browser, then get an API key for the REST endpoint.
TALENT · average rank across datasets · top 8 of 13 · lower is better
TALENT · average rank across datasets · top 8 of 13
| Rank | Entry | Average rank across datasets |
|---|---|---|
| #1 | TabICL v2 (fine-tuned) | 3.48 |
| #2 | EXAONE-Tabular | 3.85 |
| #3 | TabPFN v3.0 | 4.23 |
| #4 | TabICL v2 | 4.40 |
| #5 | TabH2O v1.1 | 5.28 |
| #6 | TabPFN v2.6 | 5.89 |
| #7 | TabPFN v2.5 | 6.23 |
| #8 | AutoGluon | 8.07 |
TabArena · average rank across datasets · top 8 of 39 · lower is better
TabArena · average rank across datasets · top 8 of 39
| Rank | Entry | Average rank across datasets |
|---|---|---|
| #1 | TabFM ensemble (official results) | 1.76 |
| #2 | TabFM (official results) | 3.31 |
| #3 | EXAONE-Tabular | 4.96 |
| #4 | TabPFN v3.0 | 6.37 |
| #5 | TabICL v2 (fine-tuned) | 7.18 |
| #6 | TabPFN v2.6 | 7.51 |
| #7 | TabH2O v1.1 | 8.39 |
| #8 | TabICL v2 | 8.76 |
Leaderboards read 25 Oct 2026. Each board ranks its own field.