# H2O-Lightning-31B

Decision models · checked 2026-11-09 · https://dbggr.dev/models/h2o-lightning-31b/

Open-weight 31B H2O.ai decision model on Gemma 4 31B for text and images; non-commercial

H2O-Lightning-31B is an open-weight decision model from H2O.ai (H2O Model Non-Commercial Licence v1.1 (non-commercial, non-production use only); gated on Hugging Face, 31B, fine-tuned from Gemma 4 31B (v1.3.4) parameters): it answers typed questions, such as yes or no, pick one or pick several, with probabilities instead of writing text. There's no hosted version: you run it on your own hardware. It scores 84.3 (v1.0.7; self-hosted; seconds are the board's estimate) on the independent JevBench Capability Score (board v1.7.46), where Jev, the reference, scores 77.1.

- Made by: H2O.ai
- Weights: Open, H2O Model Non-Commercial Licence v1.1 (non-commercial, non-production use only); gated on Hugging Face
- Size: 31B, fine-tuned from Gemma 4 31B (v1.3.4)
- Where: Self-host only
- Images: Yes
- Context: 131,072 in the Decision Index profile
- Docs: https://huggingface.co/h2oai/h2o-lightning-31b

## H2O-Lightning-31B price

US dollars per million input tokens. Decision models don't write text, so none of the hosted ones charge for output.

| Where | Input | Output | Notes | Checked |
|---|---|---|---|---|
| Self-host only | Self-host | – | – | [2026-11-09](https://huggingface.co/h2oai/h2o-lightning-31b) |

## H2O-Lightning-31B benchmarks

| Benchmark | Result | What it measures | Reported by |
|---|---|---|---|
| JevBench v1.7.46 | 84.3 | Capability Score: accuracy and calibration on typed decisions, with Jev as the reference system | [Independent](https://benchmarkheaven.com/jev-models) |
| JevBench time | 0.28 s | Seconds per decision: measured end to end for hosted APIs, the board's estimate for self-hosted models | [Independent](https://benchmarkheaven.com/jev-models) |
| Reported speed | 41.1 ms median over JevBench's 231 public items on one RTX PRO 6000 | Latency on the maker's own hardware | H2O.ai |

Makers' speeds are measured on their own hardware and aren't comparable with each other or with JevBench's timings.

## Decision models like H2O-Lightning-31B

| Model | Input / M | JevBench | Weights |
|---|---|---|---|
| H2O-Lightning-31B | Self-host | 84.3 | Open |
| [Torchcast Decision 31B](https://dbggr.dev/models/torchcast-decision-31b/) | Self-host | 91.8 | Open |
| [Piccolo-27B](https://dbggr.dev/models/piccolo-27b/) | Self-host | 86.0 | Open |
| [Mammoth-1](https://dbggr.dev/models/mammoth-1/) | $0.10 | 85.2 | Open |
| [Drex Max v1.1](https://dbggr.dev/models/drex-max-v1-1/) | $0.03 | 84.2 | Closed |
| [Perplexity Decider V1.2](https://dbggr.dev/models/pplx-decider-v1-2/) | $0.02 | 80.8 | Open |
| [d1-ultra](https://dbggr.dev/models/d1-ultra/) | $0.042 | 80.5 | Closed |
| [Kev-27B](https://dbggr.dev/models/kev-27b/) | Self-host | 78.9 | Open |
| [H2O-Lightning-4B](https://dbggr.dev/models/h2o-lightning-4b/) | Self-host | 77.0 | Open |
| [Piccolo-9B](https://dbggr.dev/models/piccolo-9b/) | Self-host | 76.6 | Open |
| [Clef](https://dbggr.dev/models/clef/) | $0.24 | 75.3 | Open |
| [Jev 1.14](https://dbggr.dev/models/jev/) | $0.042 | 75.0 | Closed |
| [d1](https://dbggr.dev/models/d1/) | $0.02 | 74.4 | Closed |
| [iota-decide-4b](https://dbggr.dev/models/iota-decide-4b/) | Self-host | 74.0 | Open |
| [Mercury Decide](https://dbggr.dev/models/mercury-decide/) | $0.02 | 73.6 | Closed |
| [Piccolo-4B](https://dbggr.dev/models/piccolo-4b/) | Self-host | 73.6 | Open |
| [OpenAI Decisions (GPT-6 Luna)](https://dbggr.dev/models/openai-decisions/) | $0.10 | 73.5 | Closed |
| [Microsoft-Decision-1](https://dbggr.dev/models/microsoft-decision-1/) | $0.042 | 70.8 | Closed |
| [Elephant-1](https://dbggr.dev/models/elephant-1/) | $0.035 | 70.2 | Open |
| [Strands Decider 26B-A4B](https://dbggr.dev/models/strands-decider-26b-a4b/) | Self-host | 66.7 | Open |
| [Clef-flash](https://dbggr.dev/models/clef-flash/) | $0.038 | 63.8 | Open |
| [Perplexity Decider Flash V1.2](https://dbggr.dev/models/pplx-decider-flash-v1-2/) | $0.01 | 61.7 | Open |
| [Kev-9B](https://dbggr.dev/models/kev-9b/) | Self-host | 61.5 | Open |
| [Clef-omni](https://dbggr.dev/models/clef-omni/) | $0.15 | 58.3 | Open |
| [Joey-1](https://dbggr.dev/models/joey-1/) | $0.01 | 57.8 | Open |
| [d1-3B](https://dbggr.dev/models/d1-3b/) | Self-host | 49.5 | Open |
| [d1-omni-600M (experimental)](https://dbggr.dev/models/d1-omni-600m/) | Self-host | 32.8 | Open |
| [Laya](https://dbggr.dev/models/laya/) | $0.011 | 32.5 | Open |
| [CLM-8B](https://dbggr.dev/models/clm-8b/) | Self-host | 20.1 | Open |
| [Clairic](https://dbggr.dev/models/clairic/) | $0.10 | – | Closed |
| [d1.1](https://dbggr.dev/models/d1-1/) | $0.02 | – | Closed |
| [d1-2B](https://dbggr.dev/models/d1-2b/) | Self-host | – | Open |
| [d1-omni-2B](https://dbggr.dev/models/d1-omni-2b/) | Self-host | – | Open |
| [Drex 1.5](https://dbggr.dev/models/drex-v1-5/) | $0.04 | – | Open |
| [Drex 2.0](https://dbggr.dev/models/drex-v2-0/) | $0.03 | – | Open |
| [Kev-0.8B](https://dbggr.dev/models/kev-0-8b/) | Self-host | – | Open |
| [Kev-4B](https://dbggr.dev/models/kev-4b/) | $0.042 | – | Open |
| [Koa Decisions](https://dbggr.dev/models/koa-decisions/) | Self-host | – | Closed |
| [Strands Decider 12B](https://dbggr.dev/models/strands-decider-12b/) | Self-host | – | Open |
| [Strands Decider 2B](https://dbggr.dev/models/strands-decider-2b/) | Self-host | – | Open |
| [Strands Decider E2B](https://dbggr.dev/models/strands-decider-e2b/) | Self-host | – | Open |
| [Strands Decider E4B](https://dbggr.dev/models/strands-decider-e4b/) | Self-host | – | Open |

## FAQ

### How much does H2O-Lightning-31B cost?

H2O-Lightning-31B has no hosted, per-token price: you download it and run it yourself, so the cost is your own compute.

### Is H2O-Lightning-31B open weight?

Yes. H2O.ai publishes the weights under H2O Model Non-Commercial Licence v1.1 (non-commercial, non-production use only); gated on Hugging Face, so you can run and fine-tune it yourself.

### How fast is H2O-Lightning-31B?

H2O.ai reports 41.1 ms median over JevBench's 231 public items on one RTX PRO 6000, measured on its own hardware. JevBench gives 0.28 seconds per decision, its estimate for self-hosting.

### How does H2O-Lightning-31B compare with Jev?

On JevBench's Capability Score (board v1.7.46), H2O-Lightning-31B scores 84.3 against Jev's 77.1. Jev costs $0.042 per million input tokens; H2O-Lightning-31B is self-host only. JevBench is a one-person project that uses Jev as its reference, which favours cheap hosted models.