Primary use stated or inferred cautiously from official documentation.
Ai2
AI2 / allenai/HiRO-ACEv2
HiRO-ACEv2
Ai2 system announced on October 14, 2026 combining the ACE2S-X-SHiELD atmospheric emulator with the HiROv2 diffusion model. It transforms roughly 100 km fields into 3 km precipitation, ten-metre wind and sea-level pressure data. The public repository includes both checkpoints, configuration and sample data; its creation on October 12 is not confused with the announcement date. It is not a conversational assistant or a conventional image generator.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | allenai/HiRO-ACEv2 |
|---|---|
| Model type | Time series |
| Access model | Open weights |
| License | Apache 2.0 |
| Deployment | Self-hosted with fme and climate data |
| Release | 14 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Numerical data |
| Exit | Numerical data |
| Context window | Climate fields at 100 km; not a token window |
| maximum output | 3 km fields at six-hour intervals |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · fme |
Climate research, simulation ensembles and extreme-event studies within conditions represented by X-SHiELD. It produces six-hour mean precipitation and instantaneous wind and pressure at that frequency. The fme package runs the emulator and downscaling with configured paths for initial conditions, forcing and outputs. The technical report describes evaluation against the reference numerical simulator.
Not suitable for operational weather forecasting: it reproduces a simulated climate, not observations. Do not extrapolate to warming scenarios or conditions outside training. HiROv2 was not trained or evaluated beyond 66 degrees south to 70 degrees north and cannot run beyond 88 degrees latitude. Joining regions may introduce discontinuities or artefacts. Realistic-looking hurricanes do not guarantee local predictive skill.
03 / CAPABILITIES
What it can do
Orientation
Climate research, simulation ensembles and extreme-event studies within conditions represented by X-SHiELD. It produces six-hour mean precipitation and instantaneous wind and pressure at that frequency. The fme package runs the emulator and downscaling with configured paths for initial conditions, forcing and outputs. The technical report describes evaluation against the reference numerical simulator.
Context
Climate fields at 100 km; not a token window · 3 km fields at six-hour intervals
Tools and integration
Not published
Access
Hugging Face · HiROv2 and ACE2S-X-SHiELD checkpoints
Input modalities
Numerical data
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
Weights under Apache 2.0, with no verified commercial endpoint or token pricing. Budget for infrastructure, scientific field storage and data preparation. Reported cost savings against X-SHiELD compare inference under Ai2’s protocol; they are not a dollar price or a guarantee for every deployment. The faster simplified version and new variables described as future work are not presented as available.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
The report and announcement describe inference costs more than 300 times lower than the reference numerical model and agreement of extreme distributions up to the 99.9999th percentile. These are author-reported results on X-SHiELD simulations, not percentage accuracy for real forecasts or general benchmarks comparable with text models. Inferama has not reproduced them. Record region, variable, period, ensemble, configuration and hardware before extrapolating savings.
i Published scientific evaluations against X-SHiELD; no general-purpose score or Inferama reproduction.
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
HiRO-ACEv2
System announced with HiROv2 and ACE2S-X-SHiELD checkpoints available.
07 / SOURCES