whatmodel.sh / NVILA-15B
NVILA-15B
↗ share
+ contribute
NVILA-15B is an open-weights language model from NVIDIA, released December 2024. Licence: cc-by-nc-4.0.
Catalogue identity notes
Comparison-only. NVIDIA / MIT Han Lab NVILA-15B vision-language model (arXiv 2412.04468; card: 'NVILA was trained in Nov 2024'), in repo Efficient-Large-Model/NVILA-15B @e621f62 (created 2024-12-04). The card's YAML licence is cc-by-nc-4.0, but its License section says the weights are CC-BY-NC-SA-4.0; both are non-commercial. Arena vision entry 'nvila-internal-15b-v1' (organization nvidia, first published 2024-12-22). Arena's own leaderboard table (arena-leaderboard Space @6f2eeeb, leaderboard_table_20250804.csv) displays that key as 'NVILA-15B' and links this repo. The key's 'internal ... v1' suggests NVIDIA served the model to Arena from its own endpoint. No run option is listed: the card gives no copy-paste command (it points to the NVLabs/VILA code), and NVIDIA offers no hosted API for it. Parameters are not stated beyond the name. Distinct from the catalogued VILA-U 7B.
Model information sources
$ how it compares
The LLMs ranked either side of NVILA-15B, out of 1132 with a score, then more from NVIDIA.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 859 |  Mixtral 8x7B v0.1 (base) | 41±1 | local | ||
| 860 |  North Micro Vision Instruct (2.4B) | 41±2 | local | ||
| 861 | NVILA-15B | 41±4 | comparison only | ||
| 862 |  CodeQwen1.5 7B (base) | 41±1 | local | ||
| 863 |  Granite 4.0 1B | 41±1 | local | ||
| more from NVIDIA | |||||
| 109 | Nemotron 3 Ultra 550B-A55B | 69±1 | api $1.35/M +local | ||
| 182 | Nemotron Cascade 2 30B-A3B | 66±1 | local | ||
| 198 | Nemotron 3 Super 120B-A12B | 65±1 | api $0.28/M +local | ||
$ what it's good at
vision & documents▁▁▃▃▂▃▅▅▇▇█▃ #188 of 199 * · 93% of its score · 1 test
Beats Gemma 3 4B, CogVLM2 Llama3 Chat 19B, MiniCPM-V 2.6 here, though they rank higher overall. Trails Qwen3.5 0.8B, Molmo 7B-D (0924) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena vision (style controlled) | 985.4 ✎ | ▁▁▃▂▂▂▄▅▅▆██ beats 7/145 | 93% |
$ where to run it
No verified API or download recorded yet · add one
Weights and documentation: huggingface.co/Efficient-Large-Model ↗
$ how the score was worked out · LMArena leaderboard dataset 93%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 LMArena leaderboard dataset | 1 | 93% |
| prior · pull toward the middle | 7% | |
| weighted average | = 40.7 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json