Project AURIONAI Use-Cases Research & Innovation

GenEval

Object-focused text-to-image compositional alignment: single object, two objects, counting, colours, position, colour attribution; scored with object detection

generative media

Score, Codesota aggregator

  1. 1SalesforceBLIP3o-NEXT-GRPO-GenEval (3B)0.91 scorevendor-reported
  2. 2Shanghai AI LabLumina-DiMOO w/ Self-GRPO0.91 scorevendor-reported
  3. 3SenseNova-U1-A3B-MoT0.91 scorevendor-reported
7scores
7models
0independent
0maintainer-run
7vendor-reported

Definition

Version
not in the notes
Category
generative media
Maintainer
GenEval authors (Ghosh, Hajishirzi, Schmidt)
Status
not in the notes
Home page
https://github.com/djghosh13/geneval

Sources

  1. github.com/djghosh13/geneval https://github.com/djghosh13/geneval
  2. arxiv.org/abs/2310.11513 https://arxiv.org/abs/2310.11513

Scores at a Glance

Score, Codesota aggregator

Higher is better for this unit.

1SalesforceBLIP3o-NEXT-GRPO-GenEval (3B)Salesforcepaper-reported, labelled unverified by Codesota; year 2025
0.91 score
vendor-reported2Shanghai AI LabLumina-DiMOO w/ Self-GRPOShanghai AI Labpaper-reported, labelled unverified by Codesota; year 2025
0.91 score
vendor-reported3SenseNova-U1-A3B-MoTpaper-reported, labelled unverified by Codesota; year 2026
0.91 score
vendor-reported4ByteDanceBAGEL (7B MoT)ByteDancepaper-reported, labelled unverified by Codesota; year 2025; with LLM rewriter
0.88 score
vendor-reported5BAAIEmu3.5BAAIpaper-reported, labelled unverified by Codesota; year 2025; 34B, AR
0.86 score
vendor-reported6SalesforceBLIP3-o (8B)Salesforcepaper-reported, labelled unverified by Codesota; year 2025
0.84 score
vendor-reported7Stanford UniversityAsymFLUX.2 kleinStanford Universitypaper-reported, labelled unverified by Codesota; year 2026
0.82 score
vendor-reported
00.250.50.751
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.

Scores

7 of 7 scores shown geneval-scores.csv
ModelScoreWho ran itSettingView or populationDateSource
SalesforceBLIP3o-NEXT-GRPO-GenEval (3B)0.91 scorevendor-reportedpaper-reported, labelled unverified by Codesota; year 2025rank 2 of 7, Codesota aggregator17 Oct 2025arxiv.org
Shanghai AI LabLumina-DiMOO w/ Self-GRPO0.91 scorevendor-reportedpaper-reported, labelled unverified by Codesota; year 2025rank 1 of 7, Codesota aggregator7 Oct 2025arxiv.org
SenseNova-U1-A3B-MoT0.91 scorevendor-reportedpaper-reported, labelled unverified by Codesota; year 2026rank 3 of 7, Codesota aggregator12 May 2026arxiv.org
ByteDanceBAGEL (7B MoT)0.88 scorevendor-reportedpaper-reported, labelled unverified by Codesota; year 2025; with LLM rewriterrank 4 of 7, Codesota aggregatorMay 2025arxiv.org
BAAIEmu3.50.86 scorevendor-reportedpaper-reported, labelled unverified by Codesota; year 2025; 34B, ARrank 5 of 7, Codesota aggregator30 Oct 2025arxiv.org
SalesforceBLIP3-o (8B)0.84 scorevendor-reportedpaper-reported, labelled unverified by Codesota; year 2025rank 6 of 7, Codesota aggregator14 May 2025arxiv.org
Stanford UniversityAsymFLUX.2 klein0.82 scorevendor-reportedpaper-reported, labelled unverified by Codesota; year 2026rank 7 of 7, Codesota aggregatorMay 2026arxiv.org
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "GenEval", https://aurion.semai.id/benchmarks/geneval/, records of October 2026.