Benchmark Almanac

MMSI-Bench

Nemotron 3.5 Super VL 43.1% leads since October 2026
Gemini 3 Pro 49.2% leads since November 2025
Gemini 3 Pro 49.2% leads since November 2025
Fading

The official leaderboard hasn’t scored a model released after Gemini 3 Pro (November 2025). Labs report their own numbers for 6 newer models, such as Nemotron 3.5 Super VL at 43.1%.

Models answer 1,000 human-written questions that need spatial reasoning across several images. Images come from indoor scans, driving, egocentric video and robot data.

Tasks cover camera, object and region relations, plus a multi-step reasoning category. arXiv v3 was posted on May 23rd, 2026.

Released
May 2025
Kind
Spatial reasoning and counting
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
Gemini 3 Pro #56 on the capability index
Last in a launch post
October 2026, dropped before it saturated
Last model added
October 2026
Scores refreshed
November 19th, 2026

In the authors’ words

Spatial intelligence is essential for multimodal large language models (MLLMs) operating in the complex physical world. Existing benchmarks, however, probe only single-image relations and thus fail to assess the multi-image spatial reasoning that real-world deployments demand. We introduce MMSI-Bench, a VQA benchmark dedicated to multi-image spatial intelligence. Six 3D-vision researchers spent more than 300 hours meticulously crafting 1,000 challenging, unambiguous multiple-choice questions from over 120,000 images, each paired with carefully designed distractors and a stepwise reasoning process. We conduct extensive experiments and evaluate 37 open-source and proprietary MLLMs, observing a wide gap: the strongest open-source model attains roughly 30% accuracy and OpenAI's GPT-5 reasoning model reaches 40%, while humans score 97%. These results underscore the challenging nature of MMSI-Bench and the substantial headroom for future research. Leveraging the annotated reasoning processes, we also provide an automated error analysis pipeline that diagnoses four dominant failure modes, including (1) grounding errors, (2) overlap-matching and scene-reconstruction errors, (3) situation-transformation reasoning errors, and (4) spatial-logic errors, offering insights for advancing spatial intelligence. Project page: https://runsenxu.com/projects/MMSI_Bench .

Abstract of MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence (arXiv:2505.23764, CC0)
7
Models scored in lab cards
3
Launch posts list it, from 3 labs
5
System cards report it, from 3 labs

Nemotron 3.5 Super VL leads with 43.1%

Best score per model as labs report them in their own cards, drawn striped. Labs pick their own settings, so these compare less cleanly than independent runs.

Gemini 3 Pro leads with 49.2% across all independent runs

Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Dotted bars are reruns by Epoch AI or another board, and each row names its source.

Gemini 3 Pro leads with 49.2%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.

  1. 1 Nemotron 3.5 Super VL 43.1%
  2. 2 SenseNova-SI-1.3-InternVL3-8B 42.5%
  3. 3 Seed2.1 Pro 35.9%
  4. 4 Seed2.0 Pro 32.5%
  5. 5 STEP3-VL-10B 32.2%
  6. 6 Seed1.8 25.8%
  7. 7 MiniCPM-V-4.7-35B-A3B 16%

Source: Labs’ model cards

Sources: Official MMSI-Bench leaderboard and labs’ model cards

#ModelScoreSettingDateSource
1 Nemotron 3.5 Super VL 43.1% Oct 2026 Lab model card
2 SenseNova-SI-1.3-InternVL3-8B 42.5% Jan 2026 Lab model card
3 Seed2.1 Pro 35.9% Jun 2026 Lab model card
4 Seed2.0 Pro 32.5% Feb 2026 Lab model card
5 STEP3-VL-10B 32.2% Jan 2026 Lab model card
6 Seed1.8 25.8% Dec 2025 Lab model card
7 MiniCPM-V-4.7-35B-A3B 16% Oct 2026 Lab model card
#ModelScoreSettingDateSource
1 Gemini 3 Pro 49.2% Nov 2025 Official leaderboard
2 GPT-5 41.9% Aug 2025 Official leaderboard
3 o3 41% Apr 2025 Official leaderboard
4 GPT-4.5 40.3% Feb 2025 Official leaderboard
5 Doubao-1.5-pro 33% Jan 2025 Official leaderboard
6 GPT-4.1 30.9% Apr 2025 Official leaderboard
7 Qwen2.5-VL-72B 30.7% Jan 2025 Official leaderboard
8 Claude 3.7 Sonnet 30.2% thinking Feb 2025 Official leaderboard
9 Seed1.5-VL 29.7% May 2025 Official leaderboard
10 InternVL2.5-2B 29% Dec 2024 Official leaderboard
11 InternVL2.5-8B 28.7% Dec 2024 Official leaderboard
12 DeepSeek-VL2-Small 28.6% Dec 2024 Official leaderboard
13 InternVL2_5-78B 28.5% Dec 2024 Official leaderboard
14 InternVL3-78B 28.5% Apr 2025 Official leaderboard
15 LLaVA-OneVision 72B 28.4% Aug 2024 Official leaderboard
16 NVILA 8B 28.1% Dec 2024 Official leaderboard
17 InternVL2.5-26B 28% Dec 2024 Official leaderboard
18 DeepSeek-VL2 27.1% Dec 2024 Official leaderboard
19 InternVL3-1B 27% Apr 2025 Official leaderboard
20 InternVL3-9B 26.7% Apr 2025 Official leaderboard
21 Qwen2.5-VL-3B 26.5% Jan 2025 Official leaderboard
22 InternVL2.5-4B 26.3% Dec 2024 Official leaderboard
23 InternVL2.5-1B 26.1% Dec 2024 Official leaderboard
24 Qwen2.5-VL-7B 25.9% Jan 2025 Official leaderboard
25 InternVL3-8B 25.7% Apr 2025 Official leaderboard
26 InternVL3-2B 25.3% Apr 2025 Official leaderboard
27 LLaVA-OneVision 7B 24.5% Aug 2024 Official leaderboard
28 DeepSeek-VL2-Tiny 24% Dec 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 3 Pro 49.2% Nov 2025 Official leaderboard
2 Nemotron 3.5 Super VL 43.1% Oct 2026 Lab model card
3 SenseNova-SI-1.3-InternVL3-8B 42.5% Jan 2026 Lab model card
4 GPT-5 41.9% Aug 2025 Official leaderboard
5 o3 41% Apr 2025 Official leaderboard
6 GPT-4.5 40.3% Feb 2025 Official leaderboard
7 Seed2.1 Pro 35.9% Jun 2026 Lab model card
8 Doubao-1.5-pro 33% Jan 2025 Official leaderboard
9 Seed2.0 Pro 32.5% Feb 2026 Lab model card
10 STEP3-VL-10B 32.2% Jan 2026 Lab model card
11 GPT-4.1 30.9% Apr 2025 Official leaderboard
12 Qwen2.5-VL-72B 30.7% Jan 2025 Official leaderboard
13 Claude 3.7 Sonnet 30.2% thinking Feb 2025 Official leaderboard
14 Seed1.5-VL 29.7% May 2025 Official leaderboard
15 InternVL2.5-2B 29% Dec 2024 Official leaderboard
16 InternVL2.5-8B 28.7% Dec 2024 Official leaderboard
17 DeepSeek-VL2-Small 28.6% Dec 2024 Official leaderboard
18 InternVL2_5-78B 28.5% Dec 2024 Official leaderboard
19 InternVL3-78B 28.5% Apr 2025 Official leaderboard
20 LLaVA-OneVision 72B 28.4% Aug 2024 Official leaderboard
21 NVILA 8B 28.1% Dec 2024 Official leaderboard
22 InternVL2.5-26B 28% Dec 2024 Official leaderboard
23 DeepSeek-VL2 27.1% Dec 2024 Official leaderboard
24 InternVL3-1B 27% Apr 2025 Official leaderboard
25 InternVL3-9B 26.7% Apr 2025 Official leaderboard
26 Qwen2.5-VL-3B 26.5% Jan 2025 Official leaderboard
27 InternVL2.5-4B 26.3% Dec 2024 Official leaderboard
28 InternVL2.5-1B 26.1% Dec 2024 Official leaderboard
29 Qwen2.5-VL-7B 25.9% Jan 2025 Official leaderboard
30 Seed1.8 25.8% Dec 2025 Lab model card
31 InternVL3-8B 25.7% Apr 2025 Official leaderboard
32 InternVL3-2B 25.3% Apr 2025 Official leaderboard
33 LLaVA-OneVision 7B 24.5% Aug 2024 Official leaderboard
34 DeepSeek-VL2-Tiny 24% Dec 2024 Official leaderboard
35 MiniCPM-V-4.7-35B-A3B 16% Oct 2026 Lab model card

The benchmark's own leaderboard has 37 entries, 28 of them models we track. Switch the scores above to rank them together.

3 models have held number one in labs’ own numbers since its release in May 2025 3 models have held number one across all independent runs since its release in May 2025 3 models have held number one with claims included since its release in May 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

OpenAI Google DeepMind ByteDance Seed SenseTime NVIDIA
Own leaderboard Not tested since Nov 2025
Gemini 3 Pro 49.2%
Lab claims 3 leaders

Sources: Official MMSI-Bench leaderboard and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Pro GPT-5 GPT-4.5 Doubao-1.5-pro LLaVA-OneVision 72B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Pro GPT-5

Sources: Official MMSI-Bench leaderboard and labs’ model cards

Labs headline it less often now, last in October 2026

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

ByteDance Seed
SenseTime
OpenBMB (ModelBest)
DateLaunch postScoreAs the post writes it
Oct 7, 2026 MiniCPM-V-4.7-35B-A3B 16 MMSI-Bench
Jan 8, 2026 SenseNova-SI-1.3-InternVL3-8B 42.5 MMSI
Dec 18, 2025 Seed1.8 25.8 MMSIBench (circular)

5 system cards report it, from 3 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Oct 15, 2026 Nemotron 3.5 Super VL technical report 43.1 MMSI, SpatialClaw harness, with agentic vision training, Table 7 Capabilities
Jun 23, 2026 Seed2.1 Pro model card 35.9 MMSIBench (circular) (Table 9) Capabilities
Mar 21, 2026 Seed1.8 model card 25.8 MMSIBench (circular) (Table 2) Capabilities
Feb 14, 2026 Seed2.0 Pro model card 32.5 MMSIBench (circular) (Table 8) Capabilities
Jan 14, 2026 STEP3-VL-10B technical report 32.2 MMSI-Bench, SeRe (Tables 1 and 3). PaCoRe 36.40 Capabilities

More in Vision

All 636 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. Qwen 3.6 Plus 86%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy. Seed2.1 Pro 90.7%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (37 entries), launch posts (3 scores) and system cards (5 scores).
  • Pass Models scored 35 models we track, plus 9 other entries on its leaderboard.
  • Pass Newest score The newest is from October 2026, for Nemotron 3.5 Super VL.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3 Pro (#56).
  • Pass Headroom The best is 49.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 28 rows for models we track and 9 other entries at the last read.
  • Pass Last read November 19th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check