Benchmark Almanac

MoI-Bench

Not in model cards

Models generate and edit images for mobile users. The authors compare image quality across 64 subtasks and measure the cost of running models on phones.

Made by
Kailai Feng and colleagues
Released
October 2026
Kind
Image generation and editing
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images → Images
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Last model added
February 2026
Scores refreshed
November 20th, 2026
19
Entries in its paper, 5 models we track

GPT-Image-2 leads with 0.97 among models released by February 2026 GPT-Image-2 leads with 0.97 among models released by February 2026

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in February 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-Image-2 GPT-4o judge 0.97
  2. 2 Seedream 5.0 Pro GPT-4o judge 0.95
  3. 3 Gemini 3.1 Flash Image GPT-4o judge 0.95
  4. 4 FLUX.2 [klein] 4B GPT-4o judge 0.83
  5. 5 DeepGen 1.0 GPT-4o judge 0.69
  6. 6 Lance-3B GPT-4o judge 0.67
  7. 7 LongCat-Image-Edit GPT-4o judge 0.67
  8. 8 Ovis-U1-3B GPT-4o judge 0.66
  9. 9 BAGEL-7B-MoT GPT-4o judge 0.64
  10. 10 DreamLite GPT-4o judge 0.62
  11. 11 OmniGen2 GPT-4o judge 0.62
  12. 12 VIBE-Image-Edit GPT-4o judge 0.59
  13. 13 InternVL-U-4B GPT-4o judge 0.51
  14. 14 Lumina-DiMOO-8B GPT-4o judge 0.43
  15. 15 LongCat-Image GPT-4o judge 0.12
  16. 16 Mobile-O-0.5B GPT-4o judge 0.10
  17. 17 SANA-600M GPT-4o judge 0.10
  18. 18 SANA-Sprint-0.6B GPT-4o judge 0.10
  19. 19 Janus-Pro-1B GPT-4o judge 0.08

Source: The MoI-Bench paper

#EntryScoreSettingDateSource
1 GPT-Image-2 0.97 GPT-4o judge Official leaderboard
2 Seedream 5.0 Pro 0.95 GPT-4o judge Official leaderboard
3 Gemini 3.1 Flash Image 0.95 GPT-4o judge Feb 2026 Official leaderboard
4 FLUX.2 [klein] 4B 0.83 GPT-4o judge Official leaderboard
5 DeepGen 1.0 0.69 GPT-4o judge Official leaderboard
6 Lance-3B 0.67 GPT-4o judge Official leaderboard
7 LongCat-Image-Edit 0.67 GPT-4o judge Official leaderboard
8 Ovis-U1-3B 0.66 GPT-4o judge Official leaderboard
9 BAGEL-7B-MoT 0.64 GPT-4o judge May 2025 Official leaderboard
10 DreamLite 0.62 GPT-4o judge Official leaderboard
11 OmniGen2 0.62 GPT-4o judge Official leaderboard
12 VIBE-Image-Edit 0.59 GPT-4o judge Official leaderboard
13 InternVL-U-4B 0.51 GPT-4o judge Official leaderboard
14 Lumina-DiMOO-8B 0.43 GPT-4o judge Official leaderboard
15 LongCat-Image 0.12 GPT-4o judge Official leaderboard
16 Mobile-O-0.5B 0.10 GPT-4o judge Official leaderboard
17 SANA-600M 0.10 GPT-4o judge Official leaderboard
18 SANA-Sprint-0.6B 0.10 GPT-4o judge Official leaderboard
19 Janus-Pro-1B 0.08 GPT-4o judge Official leaderboard
#ModelScoreSettingDateSource
1 GPT-Image-2 0.97 GPT-4o judge Official leaderboard
2 Seedream 5.0 Pro 0.95 GPT-4o judge Official leaderboard
3 Gemini 3.1 Flash Image 0.95 GPT-4o judge Feb 2026 Official leaderboard
4 FLUX.2 [klein] 4B 0.83 GPT-4o judge Official leaderboard
5 BAGEL-7B-MoT 0.64 GPT-4o judge May 2025 Official leaderboard

Rankings by part

We read 47 parts from the benchmark’s paper in 16 groups: criterion group, generation, local editing, global editing, understand. editing, ref. editing, general objects, animals, portraits, text&docs., if, vq, dp, operation, target and criteria. Each ranks models on one part of the test. GPT-Image-2 leads 41 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Instruction following

5 models we track have a score here.

  1. 1 GPT-Image-2 0.97
  2. 2 Seedream 5.0 Pro 0.95
  3. 3 Gemini 3.1 Flash Image 0.95
  4. 4 FLUX.2 [klein] 4B 0.82
  5. 5 BAGEL-7B-MoT 0.69

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-Image-2
Gemini 3.1 …
0.97
Seedream 5.0 Pro
Gemini 3.1 …
0.95
Gemini 3.1 Flash Image
Gemini 3.1 …
0.95
FLUX.2 [klein] 4B
Gemini 3.1 … GPT-4o judge
0.83
BAGEL-7B-MoT
Gemini 3.1 … GPT-4o judge
0.64

Source: The MoI-Bench paper

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-Image-2 GPT-4o judge 0.97
GPT-Image-2 Gemini 3.1 Pro judge 0.96
Seedream 5.0 Pro GPT-4o judge 0.95
Gemini 3.1 Flash Image GPT-4o judge 0.95
Seedream 5.0 Pro Gemini 3.1 Pro judge 0.93
Gemini 3.1 Flash Image Gemini 3.1 Pro judge 0.93
FLUX.2 [klein] 4B GPT-4o judge 0.83
FLUX.2 [klein] 4B Gemini 3.1 Pro judge 0.79
BAGEL-7B-MoT GPT-4o judge 0.64
BAGEL-7B-MoT Gemini 3.1 Pro judge 0.60

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-Image-2
Gemini 3.1 …
0.97
Seedream 5.0 Pro
Gemini 3.1 …
0.95
Gemini 3.1 Flash Image
Gemini 3.1 …
0.95
FLUX.2 [klein] 4B
Gemini 3.1 … GPT-4o judge
0.83
BAGEL-7B-MoT
Gemini 3.1 … GPT-4o judge
0.64

Source: The MoI-Bench paper

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-Image-2 GPT-4o judge 0.97
GPT-Image-2 Gemini 3.1 Pro judge 0.96
Seedream 5.0 Pro GPT-4o judge 0.95
Gemini 3.1 Flash Image GPT-4o judge 0.95
Seedream 5.0 Pro Gemini 3.1 Pro judge 0.93
Gemini 3.1 Flash Image Gemini 3.1 Pro judge 0.93
FLUX.2 [klein] 4B GPT-4o judge 0.83
FLUX.2 [klein] 4B Gemini 3.1 Pro judge 0.79
BAGEL-7B-MoT GPT-4o judge 0.64
BAGEL-7B-MoT Gemini 3.1 Pro judge 0.60

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Benchmark paper Not tested since Oct 2026

Source: The MoI-Bench paper

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
In the benchmark’s paper
0.60 0.70 0.80 0.90 1.00 2025 2026 0.60 0.70 0.80 0.90 1.00 2025 2026

Source: Official MoI-Bench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

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

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (19 entries).
  • Pass Models scored 5 models we track, plus 14 other entries on its leaderboard.
  • Warning Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image.
  • Note Frontier models None of its models are on the capabilities index.
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 10 rows for models we track and 14 other entries at the last read.
  • Pass Last read November 20th, 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