Benchmark Almanac

SciMMIR

CLIP-base+OCR 20.5 tops its leaderboard BLIP-2 FLAN-T5-XL 3.35 leads the models we track
Not in model cards

Models match figures and tables from scientific papers with their descriptions.

The benchmark tests retrieval from an image to text and from text to an image.

Made by
Siwei Wu and colleagues
Released
January 2024
Kind
Documents, charts and text in images
Level
Expert
Scored
Checked automatically
Answers
Prediction
Tasks
All public
Measures
Capability
Language
English
In and out
Images, Text → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Vision
Last model added
November 2023
Scores refreshed
November 20th, 2026
13
Entries in its paper, 5 models we track

CLIP-base+OCR tops the leaderboard with 20.5 BLIP-2 FLAN-T5-XL leads with 3.35 among models released by November 2023

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Table 4 lists text-to-image MRR first, so this is the full-benchmark text-to-image MRR. 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 BLIP-2 FLAN-T5-XL, at 3.35. This view shows only models we track, each at its best.

The newest model with a score here came out in November 2023, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 CLIP-base+OCR Fine-tuned 20.5
  2. 2 BLIP-base+BERT Fine-tuned 12.1
  3. 3 CLIP-base Fine-tuned 8.04
  4. 4 BLIP-base Fine-tuned 6.16
  5. 5 BLIP-2 FLAN-T5-XL Fine-tuned 3.35
  6. 6 CLIP-base+BERT Fine-tuned 2.79
  7. 7 BLIP2-OPT-2.7B Zero-shot 0.03
  8. 8 BLIP-2 FLAN-T5-XXL Zero-shot 0.03
  9. 9 LLaMA-Adapter2-7B Zero-shot 0.02
  10. 10 Kosmos-2 Zero-shot 0.01
  11. 11 llava-v1.5-7b Zero-shot 0.00
  12. 12 BLIP2-OPT-6.7B Zero-shot 0.00
  13. 13 mPLUG-Owl2 Zero-shot 0.00

Source: The SciMMIR paper

#EntryScoreSettingDateSource
1 CLIP-base+OCR 20.5 Fine-tuned Official leaderboard
2 BLIP-base+BERT 12.1 Fine-tuned Official leaderboard
3 CLIP-base 8.04 Fine-tuned Official leaderboard
4 BLIP-base 6.16 Fine-tuned Official leaderboard
5 BLIP-2 FLAN-T5-XL 3.35 Fine-tuned Feb 2023 Official leaderboard
6 CLIP-base+BERT 2.79 Fine-tuned Official leaderboard
7 BLIP2-OPT-2.7B 0.03 Zero-shot Official leaderboard
8 BLIP-2 FLAN-T5-XXL 0.03 Zero-shot Feb 2023 Official leaderboard
9 LLaMA-Adapter2-7B 0.02 Zero-shot Official leaderboard
10 Kosmos-2 0.01 Zero-shot Jun 2023 Official leaderboard
11 llava-v1.5-7b 0.00 Zero-shot Oct 2023 Official leaderboard
12 BLIP2-OPT-6.7B 0.00 Zero-shot Official leaderboard
13 mPLUG-Owl2 0.00 Zero-shot Nov 2023 Official leaderboard
#ModelScoreSettingDateSource
1 BLIP-2 FLAN-T5-XL 3.35 Fine-tuned Feb 2023 Official leaderboard
2 BLIP-2 FLAN-T5-XXL 0.03 Zero-shot Feb 2023 Official leaderboard
3 Kosmos-2 0.01 Zero-shot Jun 2023 Official leaderboard
4 llava-v1.5-7b 0.00 Zero-shot Oct 2023 Official leaderboard
5 mPLUG-Owl2 0.00 Zero-shot Nov 2023 Official leaderboard

Number one over time

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

Salesforce
Benchmark paper Not tested since Jan 2024

Source: The SciMMIR 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
0.00 1.00 2.00 3.00 4.00 2023 2024 0.00 1.00 2.00 3.00 4.00 2023 2024

Source: Official SciMMIR 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 3 years ago, in November 2023, for mPLUG-Owl2. 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 (13 entries).
  • Pass Models scored 5 models we track, plus 8 other entries on its leaderboard.
  • Warning Newest score The newest came 3 years ago, in November 2023, for mPLUG-Owl2.
  • 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 aclanthology.org: 6 rows for models we track and 8 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