whatmodel.sh / InternVL2-26B

InternVL2-26B

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

InternVL2-26B is an open-weights language model from Shanghai AI Laboratory (OpenGVLab), released July 2024. Its published size is 25.5B parameters. Licence: mit.

Catalogue identity notes

Comparison-only. Arena entry 'internvl2-26b' (vision). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'internvl2-26b' as 'InternVL2-26B' (OpenGVLab, MIT), linking https://internvl.github.io/blog/2024-07-02-InternVL-2.0/. Maker repo OpenGVLab/InternVL2-26B @46f3797: licence tag mit. 25,514,186,112 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2024-07: InternVL 2.0 blog post internvl.github.io/blog/2024-07-02-InternVL-2.0/ (linked by LMArena's table).

46.8±1.5#402 of 719 LLMsfrom 17 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of InternVL2-26B, out of 828 with a score, then more from Shanghai AI Laboratory (OpenGVLab).

#modelscorewhereRAMreleased
477DeepSeek R1 Distill Llama 8B DeepSeek47±1local~6.2GB2025-01
478EXAONE 4.0 1.2B LG AI Research47±2local~2.2GB2025-07
479InternVL2-26B Shanghai AI Laboratory (OpenGVLab)47±2local~55GB2024-07
480speechless-coder-ds-6.7b uukuguy47±1local~5.8GB2023-12
481Gemma 4 E2B Google47±1api $0.050/M +local~5.1GB2026-04
more from Shanghai AI Laboratory (OpenGVLab)
604InternVL2-4B Shanghai AI Laboratory (OpenGVLab)41±2local~10GB2024-07

see all LLMs · find the ones that fit your machine

$ what it's good at

math▁▂▃▅▆▅▄█▄▁▁▁ #259 of 398 * · 6.3% of its score · 1 test

Beats Molmo 2 8B, Llama 3.3 70B Instruct, DeepSeek-V2.5 here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Mixtral 8x7B v0.1 (base), Tulu 2 DPO 70B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
MathVista (testmini) lab59.4% ✎▁··▁▄▄▇█▅▆█▄ beats 15/47medium · avg 66%▇▇▇ 6.3%at its level · small run
vision & documents▁▁▃▃▁▃▄▂▆▇█▄ #126 of 147 · 89% of its score · 14 tests

Beats Nova Pro, Nova Lite here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
CharXiv descriptive questions62.4% ✎▄█▄▄▄·██·▄·█ beats 4/12easy · avg 71%▇▇▇▇▇▇▇▇▇▇ 26%at its level
Arena vision (style controlled)1024.8 ✎▁▁▂▂▃▂▄▅▅▆██ beats 16/145avg 1068▇▇▇▇▇▇ 14%blunt
CharXiv (reasoning questions)33.4% ✎▂▂▂█▃▂·▂▂··▂ beats 7/13medium · avg 36%▇▇▇▇▇ 13%at its level
CapArena-Auto detailed image captioning (Score_Avg, GPT-4o judge)13.0000 ✎▃▃▃·▃█·▅▃▃▃▃ beats 8/12avg -0.671▇▇▇▇ 10%typical
MMMU (val) lab51.2% ✎▁····▁▁▄█▇▆▆ beats 11/52medium · avg 58%▇▇▇ 6.7%at its level · small run
MathVista (testmini) lab59.4% ✎▁··▁▄▄▇█▅▆█▄ beats 15/47medium · avg 66%▇▇▇ 6.3%at its level · small run
PhysBench physical-world understanding (test, Avg)43.5% ✎▅▃█··▅▃····▃ beats 7/9hard · avg 43%▇ 2.7%blunt
HallusionBench lab50.7% ✎▃▁·▃▃▃▄▅▄█▅▅ beats 4/32medium · avg 56%▇ 2.3%blunt · at its level · small run
6 more tests
testresultvs the others in its rundifficultycountswhy
DocVQA lab92.9% ✎▁··▃▁▁▁▇█▇▄▅ beats 7/28very easy · avg 93%▇ 2.1%blunt · maxed out · small run
ChartQA lab84.9% ✎▁··▁▁▁▁▁▂▅▅█ beats 24/44easy · avg 82%▇ 2.0%blunt · small run
RealWorldQA lab68.3% ✎▂▃▂█▃██▆█▂▆▆ beats 9/38easy · avg 71%▇ 1.7%blunt · at its level · small run
OCRBench lab82.5% ✎▁·▃▁▆▅▂█▅█▁▃ beats 13/37easy · avg 83%▇ 1.4%blunt · small run
AI2D test lab84.5% ✎▁▁▅▇▃▇·▄·█·▃ beats 16/28easy · avg 84%▇ 0.8%blunt · small run
TextVQA lab82.3% ✎▁▁▁▄▅▁▃·▃▃▃█ beats 16/24easy · avg 78%▇ 0.8%blunt · small run
other tests3 not in a task · 10% of its score
testresultvs the others in its rundifficultycountswhy
Video-MME (with subtitles) lab57.1% ✎▁▃▁▁▁·▁▃▁▁▁█ beats 1/16medium · avg 68%▇▇ 4.6%at its level · small run
Video-MME (without subtitles) lab54.8% ✎▁▃▁▃▁·█▄▄▅·▇ beats 1/27medium · avg 64%▇ 3.3%blunt · at its level · small run
MVBench lab67.5% ✎▁··▃▄▁▃▄▄▄█▄ beats 7/26medium · avg 67%▇ 1.7%blunt · at its level · small run

$ where to run it

Other ways to run it

  1. Run it as published · needs ~51.03 GB, NVIDIA only · from github.com/vllm-project
    vllm serve OpenGVLab/InternVL2-26B --trust-remote-code

Official weights and documentation: huggingface.co/OpenGVLab ↗

$ how the score was worked out · CharXiv leaderboard (princeto… 39%, OpenGVLab InternVL2 model car… 34%, LMArena leaderboard dataset 14% 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

sourceresultsshare of the scorelicence
CharXiv leaderboard (princeton-nlp/CharXiv README)2▇▇▇▇▇▇▇▇▇▇▇▇ 39%Apache-2.0
OpenGVLab InternVL2 model cards (InternVL2-4B and InternVL2-26B), Evaluation tables · lab-reported12▇▇▇▇▇▇▇▇▇▇ 34%MIT
LMArena leaderboard dataset1▇▇▇▇ 14%CC BY 4.0
CapArena: Benchmarking and Analyzing Detailed Image Captioning in the LLM Era (arXiv:2503.12329v1)1▇▇▇ 10%CC-BY-4.0
PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (arXiv:2501.16411v2)1▇ 3%CC-BY-4.0
prior · pull toward the middle▇ 1%
weighted average= 46.8

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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