Benchmark Almanac

MarsRetrieval

WeMM-Embedding-9B 45.4 tops its leaderboard EmbeddingGemma 2 26.5 leads the models we track
Not in model cards

Models match Mars images to captions, retrieve examples of named landforms, and locate features in a global mosaic of the planet.

An MTEB contributor tested SigLIP 2 and Qwen3-VL-Embedding on 2,247 of the original 2,287 image-caption pairs. The run covered both retrieval directions.

Released
February 2026
Kind
Spatial reasoning and counting
Level
University
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
Last model added
October 2026
Scores refreshed
November 20th, 2026
19
Entries in its paper, 3 models we track

WeMM-Embedding-9B tops the leaderboard with 45.4 EmbeddingGemma 2 leads with 26.5 among models released by October 2026

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports no overall score. The headline is text-to-image Recall@1 on the paired retrieval task. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is EmbeddingGemma 2, at 26.5. This view shows only models we track, each at its best.

The newest model with a score here came out in October 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 WeMM-Embedding-9B 45.4
  2. 2 Ovis-VL-Embedding-9B 41.7
  3. 3 WeMM-Embedding-2B 36.4
  4. 4 UEmbed-9B 35.6
  5. 5 Ovis-VL-Embedding-2B 35.5
  6. 6 MarScope 35.2
  7. 7 PE-Core-L-14-336 27.1
  8. 8 UEmbed-2B 26.7
  9. 9 EmbeddingGemma 2 26.5
  10. 10 Qwen3-VL-Embedding-8B 26.5
  11. 11 ViT-L-16-SigLIP2-512 25.9
  12. 12 Ops-MM-embedding-v1-7B 24.1
  13. 13 Seed1.6-Embedding-1215 23.1
  14. 14 Qwen3-VL-Embedding-2B 22.3
  15. 15 ViT-L-16-SigLIP-384 21.6
  16. 16 Ops-MM-embedding-v1-2B 21.0
  17. 17 DFN2B-CLIP-ViT-L-14 17.3
  18. 18 jina-embeddings-v4 17.3
  19. 19 aimv2-large-patch14-224 10.1

Source: The MarsRetrieval paper

#EntryScoreSettingDateSource
1 WeMM-Embedding-9B 45.4 Official leaderboard
2 Ovis-VL-Embedding-9B 41.7 Official leaderboard
3 WeMM-Embedding-2B 36.4 Official leaderboard
4 UEmbed-9B 35.6 Official leaderboard
5 Ovis-VL-Embedding-2B 35.5 Official leaderboard
6 MarScope 35.2 Official leaderboard
7 PE-Core-L-14-336 27.1 Official leaderboard
8 UEmbed-2B 26.7 Official leaderboard
9 EmbeddingGemma 2 26.5 Oct 2026 Official leaderboard
10 Qwen3-VL-Embedding-8B 26.5 Jan 2026 Official leaderboard
11 ViT-L-16-SigLIP2-512 25.9 Official leaderboard
12 Ops-MM-embedding-v1-7B 24.1 Official leaderboard
13 Seed1.6-Embedding-1215 23.1 Official leaderboard
14 Qwen3-VL-Embedding-2B 22.3 Jan 2026 Official leaderboard
15 ViT-L-16-SigLIP-384 21.6 Official leaderboard
16 Ops-MM-embedding-v1-2B 21.0 Official leaderboard
17 DFN2B-CLIP-ViT-L-14 17.3 Official leaderboard
18 jina-embeddings-v4 17.3 Official leaderboard
19 aimv2-large-patch14-224 10.1 Official leaderboard
#ModelScoreSettingDateSource
1 EmbeddingGemma 2 26.5 Oct 2026 Official leaderboard
2 Qwen3-VL-Embedding-8B 26.5 Jan 2026 Official leaderboard
3 Qwen3-VL-Embedding-2B 22.3 Jan 2026 Official leaderboard

Rankings by part

We read 8 parts from the benchmark’s paper, each ranking models on one part of the test, such as Text to image R@1, Text to image R@10 and Text to image MRR. EmbeddingGemma 2 leads 4 of them, and Qwen3-VL-Embedding-8B leads the other 4. Arrows show how far a model moved from its overall rank among the models we track.

Text to image R@1

3 models we track have a score here.

  1. 1 EmbeddingGemma 2 26.5
  2. 2 Qwen3-VL-Embedding-8B 26.5
  3. 3 Qwen3-VL-Embedding-2B 22.3

2 models have held number one in the benchmark’s paper since its release in February 2026

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

Qwen Google DeepMind
Benchmark paper Not tested since Oct 2026

Source: The MarsRetrieval 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
22.0 23.0 24.0 25.0 26.0 27.0 2026 22.0 23.0 24.0 25.0 26.0 27.0 2026

Source: Official MarsRetrieval 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

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (19 entries).
  • Pass Models scored 3 models we track, plus 16 other entries on its leaderboard.
  • Pass Newest score The newest is from October 2026, for EmbeddingGemma 2.
  • 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: 3 rows for models we track and 16 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