Benchmark Almanac

InternVL3.5-241B-A28B

InternVL3.5-241B-A28B ranks first on 9 of the 63 benchmarks with five or more models scored.

Shanghai AI Lab Released August 25th, 2025 Family InternVL3.5 The lab's model card Compare with others
76
Benchmarks with a score
9
Ranked first, out of five or more models
15
In the top three
Shanghai AI Lab Aug 2025

InternVL3.5-241B-A28B

Open weights Flagship

134.2 #235 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • ERQA 46.8% #31 /36
  • HallusionBench 57.3% #26 /31
  • MVBench 76.5% #5 /23

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
HellaSwag 97% #1 of 53 Lab model card
WinoGrande 91.7% #1 of 47 Lab model card
Natural Questions 50.4% #2 of 20 Lab model card
RACE 94.2% #1 of 8 Lab model card
POPE 90.7% #1 of 7 Lab model card
MMLU 89.1% #20 of 131 Lab model card
C3 97.8% #1 of 6 Lab model card
CMMLU 90.2% #4 of 17 Lab model card
BIG-Bench Hard 86.5% #18 of 73 Lab model card
C-Eval 90.9% #15 of 46 Lab model card
MMLU-Pro 81.3% #67 of 167 Lab model card
GAOKAO-Bench 94.5% #1 of 2 Lab model card
IFEval 83.7% #91 of 135 Lab model card
TriviaQA 74.8% #27 of 40 Lab model card
HallusionBench 57.3% #26 of 31 Lab model card

Math and science

BenchmarkScoreRankSource
MATH 94.5% #5 of 99 Lab model card
MATH-500 98% #9 of 70 Lab model card
AIME 2024 84.7% #32 of 88 Lab model card
GSM8K 91.6% #37 of 92 Lab model card
GPQA Diamond 73.2% #159 of 302 Lab model card
OlympiadBench 68.7% #2 of 3 Lab model card
AIME 2025 75.6% #101 of 143 Lab model card

Coding

BenchmarkScoreRankSource
HumanEval 98.2% #2 of 101 Lab model card

Agents and tools

BenchmarkScoreRankSource
ScreenSpot 89.8% #3 of 11 Lab model card
ScreenSpot-v2 92.9% #6 of 14 Lab model card
OSWorld-G 53.2% #4 of 5 Lab model card
Windows Agent Arena 18% #5 of 5 Lab model card

Vision, audio and robotics

BenchmarkScoreRankSource
SpaCE-10 55% #1 of 36 Official leaderboard
TextVQA 84.5% #2 of 17 Lab model card
MM-Vet 81.2% #1 of 8 Lab model card
RefCOCO / RefCOCO+ / RefCOCOg 92.4% #4 of 29 Lab model card
VSI-Bench 69.5% #1 of 7 Lab model card
MME 2525.9 #1 of 6 Lab model card
OCRBench 90.7% #7 of 34 Lab model card
MVBench 76.5% #5 of 23 Lab model card
MME-RealWorld 65.1% #1 of 4 Lab model card
MMMU 77.7% #31 of 103 Lab model card
ChartQA 88% #12 of 37 Lab model card
DocVQA 94.9% #12 of 36 Lab model card
OVO-S-Bench 50.9% #5 of 15 Official leaderboard
VCR (Visual Caption Restoration) 97% #1 of 3 Lab model card
MathVista 82.7% #25 of 70 Lab model card
GSM8K-V 27.9% #9 of 24 Official leaderboard
InfographicVQA 82% #10 of 26 Lab model card
MMStar 77.9% #14 of 36 Lab model card
VideoMME v2 (with subtitles) 23.1% #15 of 31 Official leaderboard
LogicVista 66.7% #7 of 14 Lab model card
MTVQA-Test 39.3% #3 of 6 Lab model card
OmniSpatial 51.9% #3 of 6 Lab model card
CRPE 78% #2 of 4 Lab model card
MMT-Bench 72.7% #1 of 2 Lab model card
R-Bench (real-world robustness) 75.4% #1 of 2 Lab model card
RealWorldQA 75.2% #22 of 38 Lab model card
AI2D 87.3% #33 of 52 Lab model card
MMBench-EN v1.1 87.4% #15 of 23 Lab model card
MathVerse 68.5% #8 of 12 Lab model card
SEED-Bench-2-Plus 71.9% #2 of 3 Lab model card
BLINK 61.4% #20 of 29 Lab model card
MATH-Vision (MathVision) 63.9% #36 of 51 Lab model card
MLVU 78.2% #16 of 22 Lab model card
BabyVision 19.2% #20 of 27 Official leaderboard
ViVerBench 67.1% #6 of 8 Official leaderboard
Video-MME 72.9% #40 of 51 Lab model card
LongVideoBench 67.1% #9 of 11 Lab model card
ERQA 46.8% #31 of 36 Lab model card
OmniDocBench v1.5 82.7 #10 of 11 Official leaderboard
DynaMath 46.5% #21 of 23 Lab model card
MuirBench 47.5% #11 of 12 Lab model card
We-Math 62.3% #12 of 13 Lab model card
CharXiv 48.3% #72 of 75 Lab model card
OmniDocBench v1.6 83.8 #8 of 8 Official leaderboard
Mantis-Eval 77% #2 of 2 Lab model card
MMBench-Video 1.74 #2 of 2 Lab model card
MIRB 73% #1 of 1 Lab model card
MMIU 61.3% #1 of 1 Lab model card
SGP-Bench 70.7% #1 of 1 Lab model card

Its lab's numbers put it at 134.2, number 235 of 404

The Self-Reported Index fits one score to the 45 capability results Shanghai AI Lab reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 132.4 to 136.0. How we compute it.

Its lab's numbers next to 2 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark Lab says Board says Gap
MATH-Vision (MathVision) 63.9% MathVision 63.9% Leaderboard Same
SpaCE-10 55% SpaCE-10 55% Leaderboard Same

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

InternVL3.5-241B-A28B model card (Hugging Face) Aug 25, 2025

Its system card

Every result in the longer documents the lab published with it. 35 of them aren't in the launch post.

InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency Aug 25, 2025

Capabilities

Hallucination

Robustness