Model

InternVL3-78B

OPEN

OpenGVLab (Shanghai AI Lab) · InternVL3 family · released Apr 11, 2025

Flagship of the natively-multimodal InternVL3 series, surpassing GPT-4o on MMMU; adds tool use, GUI agents and 3D perception.

ReasoningCodingVisionFunction callingTool useAgentic

1541.5

Elo · unrated

Parameters

78B

Active params

78B (dense)

Context

32K tokens

Architecture

InternViT-6B-448px-V2.5 vision encoder + Qwen2.5-72B language model, natively multimodal pre-training

License

Qwen License (InternViT component MIT; Qwen2.5-72B component under Qwen License)

Languages

—

API price (in/out)

No hosted API

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AI2DVision96.0%#2

best: Molmo 72B · 96.3%

ChartQAVision89.7%#5

best: MiniMax-VL-01 · 91.7%

CharXivVision46.0%#32

best: Muse Spark 1.1 (xhigh) · 88.4%

DocVQAVision95.4%#6

best: Qwen2-VL-72B · 96.5%

MathVisionVision43.1%#14

best: Seed 2.1 Pro · 92.6%

MathVistaVision79.0%#14

best: Seed 2.1 Pro · 90.7%

MMBench (Chinese)Vision88.7%#2

best: ERNIE 4.5 VL 424B-A47B · 90.9%

MMBench (English)Vision89.0%#7

best: Qwen3.5-397B-A17B · 93.7%

MMMUVision72.2%#47

best: Claude Fable 5 · 89.3%

OCRBenchVision906#1

best: this model · 906

RefCOCO+Vision90.1%#3

best: DeepSeek-VL2 · 91.2%

RefCOCOVision93.4%#4

best: Qwen3.5-Omni-Plus · 95.0%

TextVQAVision84.3%#4

best: Molmo 2 8B · 85.7%

Video-MMEVision72.7%#14

best: Seed 2.1 Pro · 89.2%

Run it locally

VRAM @ Q4

53 GB

VRAM @ FP16

156 GB

Fits on (Q4)

M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

78B model exceeds a single RTX 4090's 24GB VRAM even at Q4 (AWQ build is 53GB); no single-4090 measurement exists.

Quantizations

AWQ

InternVL3 family

Elo progression across releases

InternVL3-78BApr 20251541.5

Resources

HFOpenGVLab/InternVL3-78B GHOpenGVLab/InternVL APIAvailable via OpenGVLab (Shanghai AI Lab) API

API price weights · each benchmark row carries its own source badge (see methodology)