Meta-Dataset
Models learn from episodes sampled across diverse image datasets and classify novel classes from a few examples.
The official repository reports 804 GitHub stars and documents the original and VTAB plus Meta protocols.
- Made by
- Google Research and collaborators
- Released
- March 2019
- Kind
- General visual questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- New versions replace it
- Run by
- Anyone
- In views
- Vision
DIPA tops the leaderboard with 81.7%
Best score per entry from the benchmark's paper, read on November 21st, 2026. A paper’s table doesn’t add newer models. The repository reports no overall score, so this is the plain mean of its ten dataset accuracies. 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.
- 81.7 DIPA
- 78.1 TSA
- 75.8 URL
- 74.9 TSA_resnet34
- 74.3 CTX
- 72.2 URT
- 71.8 FLUTE
- 71.2 TriM
- 70.7 SUR
- 70.7 TSA_resnet18
- 70.3 TransductiveCNAPS
- 69.9 SimpleCNAPS
- 69.6 URT-pf
- 69.3 SUR-pnf
- 66.9 CNAPs
- 63.4 fo-Proto-MAML
- 61.4 ALFA+fo-Proto-MAML
- 60.6 ProtoNet
- 60.4 ProtoNet (large)
- 59.1 BOHB
- 58.8 Finetune
- 57.0 fo-MAML
- 56.3 MatchingNet
- 54.3 k-NN
- 53.3 RelationNet
- 1 DIPA training on all datasets 81.7%
- 2 TSA training on all datasets 78.1%
- 3 URL training on all datasets 75.8%
- 4 TSA_resnet34 training on ImageNet only 74.9%
- 5 CTX training on ImageNet only 74.3%
- 6 URT training on all datasets 72.2%
- 7 FLUTE training on all datasets 71.8%
- 8 TriM training on all datasets 71.2%
- 9 SUR training on all datasets 70.7%
- 10 TSA_resnet18 training on ImageNet only 70.7%
- 11 TransductiveCNAPS training on all datasets 70.3%
- 12 SimpleCNAPS training on all datasets 69.9%
- 13 URT-pf training on all datasets 69.6%
- 14 SUR-pnf training on all datasets 69.3%
- 15 CNAPs training on all datasets 66.9%
- 16 fo-Proto-MAML training on all datasets 63.4%
- 17 ALFA+fo-Proto-MAML training on ImageNet only 61.4%
- 18 ProtoNet training on all datasets 60.6%
- 19 ProtoNet (large) training on ImageNet only 60.4%
- 20 BOHB training on ImageNet only 59.1%
- 21 Finetune training on ImageNet only 58.8%
- 22 fo-MAML training on all datasets 57%
- 23 MatchingNet training on all datasets 56.3%
- 24 k-NN training on ImageNet only 54.3%
- 25 RelationNet training on all datasets 53.3%
Source: The Meta-Dataset paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | DIPA | 81.7% | training on all datasets | Official leaderboard | |
| 2 | TSA | 78.1% | training on all datasets | Official leaderboard | |
| 3 | URL | 75.8% | training on all datasets | Official leaderboard | |
| 4 | TSA_resnet34 | 74.9% | training on ImageNet only | Official leaderboard | |
| 5 | CTX | 74.3% | training on ImageNet only | Official leaderboard | |
| 6 | URT | 72.2% | training on all datasets | Official leaderboard | |
| 7 | FLUTE | 71.8% | training on all datasets | Official leaderboard | |
| 8 | TriM | 71.2% | training on all datasets | Official leaderboard | |
| 9 | SUR | 70.7% | training on all datasets | Official leaderboard | |
| 10 | TSA_resnet18 | 70.7% | training on ImageNet only | Official leaderboard | |
| 11 | TransductiveCNAPS | 70.3% | training on all datasets | Official leaderboard | |
| 12 | SimpleCNAPS | 69.9% | training on all datasets | Official leaderboard | |
| 13 | URT-pf | 69.6% | training on all datasets | Official leaderboard | |
| 14 | SUR-pnf | 69.3% | training on all datasets | Official leaderboard | |
| 15 | CNAPs | 66.9% | training on all datasets | Official leaderboard | |
| 16 | fo-Proto-MAML | 63.4% | training on all datasets | Official leaderboard | |
| 17 | ALFA+fo-Proto-MAML | 61.4% | training on ImageNet only | Official leaderboard | |
| 18 | ProtoNet | 60.6% | training on all datasets | Official leaderboard | |
| 19 | ProtoNet (large) | 60.4% | training on ImageNet only | Official leaderboard | |
| 20 | BOHB | 59.1% | training on ImageNet only | Official leaderboard | |
| 21 | Finetune | 58.8% | training on ImageNet only | Official leaderboard | |
| 22 | fo-MAML | 57% | training on all datasets | Official leaderboard | |
| 23 | MatchingNet | 56.3% | training on all datasets | Official leaderboard | |
| 24 | k-NN | 54.3% | training on ImageNet only | Official leaderboard | |
| 25 | RelationNet | 53.3% | training on all datasets | Official 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
- 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 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the benchmark’s paper (25 entries).
- Pass Models scored 25 entries on its leaderboard, none a model we track.
- Note Newest score No scores yet.
- Note Frontier models No models scored yet.
- Note Headroom No scores yet.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 25 entries at the last read, none a model we track.
- Pass Last read November 21st, 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.