whatmodel.sh / LLaVA-v1.6-34B
LLaVA-v1.6-34B
↗ share
+ contribute
LLaVA-v1.6-34B is an open-weights language model from LLaVA, released January 2024. Its published size is 34.8B parameters and a 4k-token context window. Licence: apache-2.0.
Catalogue identity notes
Arena entry 'llava-v1.6-34b' (vision). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'llava-v1.6-34b' as 'LLaVA-v1.6-34B' (LLaVA, Apache 2.0), linking https://llava-vl.github.io/blog/2024-01-30-llava-next/. Maker repo liuhaotian/llava-v1.6-34b @6754aa8: licence tag apache-2.0. 34,751,166,464 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2024-01: LLaVA-NeXT blog post llava-vl.github.io/blog/2024-01-30-llava-next/ (linked by LMArena's table).
$ how it compares
The LLMs ranked either side of LLaVA-v1.6-34B, out of 828 with a score.
| # | model | score | where |
|---|---|---|---|
| 627 |  LFM2 1.2B | 40±2 | local |
| 628 |  Code Tulu 2 13B | 40±2 | local |
| 629 | LLaVA-v1.6-34B | 40±2 | local |
| 630 |  Bielik 1.5B v3.0 Instruct | 40±3 | local |
| 631 |  OpenHermes 2.5 Mistral 7B | 39±1 | local |
$ what it's good at
vision & documents▁▃▃▂▁▃▄▂▆▇█▄ #143 of 147 · 99% of its score · 8 tests
Beats Gemma 3 4B, Phi-3-vision-128k-instruct here, though they rank higher overall. Trails Qwen3.5 0.8B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CharXiv descriptive questions | 51.0% ✎ | ▄█▄▄▄·██·▄·█ beats 0/12 | 31% |
| CharXiv descriptive questions (test split) | 51.5% ✎ | █··▄··█··▄·▄ beats 0/6 | 18% |
| Arena vision (style controlled) | 964.4 ✎ | ▁▁▃▂▂▂▄▅▅▆██ beats 5/145 | 16% |
| CapArena-Auto detailed image captioning (Score_Avg, GPT-4o judge) | -67.5000 ✎ | ▃▃▃·▃█·▅▃▃▃▃ beats 0/12 | 11% |
| CharXiv (reasoning questions) | 22.5% ✎ | ▃▂▂█▃▂·▂▂··▂ beats 0/13 | 10% |
| MMMU (val) lab | 51.1% ✎ | ▁····▁▁▄█▇▆▆ beats 10/52 | 6.8% |
| CharXiv reasoning questions (test split) | 20.0% ✎ | ▄··▄▄▄█····▄ beats 0/6 | 6.0% |
| TextVQA lab | 69.5% ✎ | ▁▁▁▄▅▁▃·▃▃▃█ beats 6/24 | 0.9% |
$ where to run it
Ollama · recommended
version
- Install Ollama
download it (macOS / Windows) ↗
Linuxcurl -fsSL https://ollama.com/install.sh | sh - Run LLaVA-v1.6-34B · needs ~21.38 GB · from ollama.comchat in the terminal
ollama run llava:34b-v1.6-q4_K_M
Weights and documentation: huggingface.co/liuhaotian ↗
$ how the score was worked out · CharXiv leaderboard (princeto… 42%, CharXiv: Charting Gaps in Rea… 24%, LMArena leaderboard dataset 16% and 2 more
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 |
|---|---|---|
 CharXiv leaderboard (princeton-nlp/CharXiv README) | 2 | 42% |
 CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs (arXiv:2406.18521v1) | 2 | 24% |
 LMArena leaderboard dataset | 1 | 16% |
| CapArena: Benchmarking and Analyzing Detailed Image Captioning in the LLM Era (arXiv:2503.12329v1) | 1 | 11% |
| LLaVA repository model zoo (haotian-liu/LLaVA docs/MODEL_ZOO.md) · lab-reported | 2 | 8% |
| prior · pull toward the middle | 1% | |
| weighted average | = 39.5 |
$ 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