ImageJevBench v0.3.0
Version 0.3.0
Models make decisions about images and report probabilities. This version draws 300 public and 1,200 sealed items from a new pool.
Benchmark Heaven’s core results list 68 measured systems and accept 65 for ranking. The maker keeps older scores and computer use separate.
- Made by
- Benchmark Heaven
- Released
- October 2026
- Kind
- General visual questions
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- New versions replace it
- Run by
- A host
- Based on
- A version of another benchmark
- In views
- Vision
- Strongest model tested
-
Gemini 3.8 Flash #24 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
kettei-fat (Sqwish) tops the leaderboard with 75.4 GPT-6 Luna leads with 17.3 among models released by September 2026
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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 GPT-6 Luna, at 17.3. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6.1 Sol
- and 16 more
- 75.4 kettei-fat (Sqwish)
- 71.2 Wity-1 (AlphaNimble), reasoning auto
- 66.6 H2O-Lightning-4B v1.4.1 (image)
- 63.7 Imajev-4B
- 63.3 kettei-chub (Sqwish)
- 51.2 H2O-Lightning-31B v1.0.7 (image)
- 51.0 H2O-Lightning-31B v1.3.4 @c35ed950 (image)
- 47.9 Sage 1.4 + images, beta (Levanto Labs)
- 44.6 Drex Max (Nace AI)
- 42.5 Surogate Rune 26B-A4B v3
- 41.4 Autoloops – Gemma 4 31B IT
- 40.9 basal-omni-1.0-ultra (Remek Kinas)
- 39.5 Celeris-1 Decision (Celeris)
- 34.5 imajev 9B
- 34.1 H2O-Lightning-4B (image)
- 33.1 Wity-1 (AlphaNimble), reasoning off
- 32.8 Jev-Omni
- 31.4 JPT-9B (kirp / llm2jev)
- 30.4 NeoHorse Jev 4B
- 29.7 Jevify Gemma 4 26B-A4B
- 29.3 AutoJev-27B
- 26.6 JPT-4B (kirp / llm2jev)
- 26.4 Jev 1.13.0 + images, preview (TypeSafe)
- 25.5 Jev-Vision 8B (SeanLiu)
- 24.3 Jev 1.14.0 + images, preview (TypeSafe)
- 22.8 shisa-de-1
- 20.4 Clef-Flash (Cloudflare)
- 18.6 Clef (Cloudflare)
- 17.3 GPT-6 Luna
- 17.1 Winnow-12B (Q8_0 + F16 vision projector)
- 15.3 Image Hopper (HopitAI)
- 15.0 OmniJev-Qwen3.5-9B-v4 (tzcfly)
- 14.8 OmniJev 4B (tinnel123)
- 13.8 Reflex 4B (released stable configuration)
- 12.0 Bonsai-2-27B v2 PQ2_0 + Q8_0 MMProj
- 11.5 CUA-S1 4B
- 11.1 Visual-Jev generic (zero-shot Qwen3.5-4B baseline)
- 9.52 Mapika decider-2b-vision BF16
- 8.86 vjev-vision (yah01)
- 8.71 Visual-Jev 4B Answer-SFT
- 7.87 GPT-5.6 Luna
- 7.26 Jevify Gemma 4 E4B
- 7.17 JPT-0.8B (kirp / llm2jev)
- 6.85 Glance (frozen Qwen3-VL-4B)
- 6.45
Gemini 3.1 Flash-Lite
- 5.90 diffusiongemma-26b djev v10 step160 (snowicarus)
- 5.76 imajev 2B
- 4.14 Gevva E4B (text checkpoint, image input)
- 4.10 Liquid d1-3B (Liquid AI)
- 3.62 djev-distill-v4 (tarsur385)
- 3.29 Jevify Qwen3-VL-2B T2
- 3.17 OmniJev 2B (tinnel123)
- 2.02 Gevva E2B multimodal
- 1.97 jev-spatial (Fr0zencr4nE)
- 1.84 JEVision (divyanshx11, visual route)
- 1.57 OmniJev 0.8B (tinnel123)
- 0.50 PlayJev 0.8B
- 0.24 Liquid d1-omni-600M (Liquid AI)
- 0.04 djev-dev BF16
- 0.03 Qevi-2B (MeerDevelopment)
- 0.00 djev-spark NVFP4
- 0.00
Gemini 3.8 Flash
- 0.00 OpenJev 4B NLI v2 (official image-premise path)
- 0.00 OpenJev 4B NLI v5
- 1 kettei-fat (Sqwish) 75.4
- 2 Wity-1 (AlphaNimble), reasoning auto auto 71.2
- 3 H2O-Lightning-4B v1.4.1 (image) 66.6
- 4 Imajev-4B 63.7
- 5 kettei-chub (Sqwish) 63.3
- 6 H2O-Lightning-31B v1.0.7 (image) 51.2
- 7 H2O-Lightning-31B v1.3.4 @c35ed950 (image) 51.0
- 8 Sage 1.4 + images, beta (Levanto Labs) 47.9
- 9 Drex Max (Nace AI) 44.6
- 10 Surogate Rune 26B-A4B v3 42.5
- 11 Autoloops – Gemma 4 31B IT 41.4
- 12 basal-omni-1.0-ultra (Remek Kinas) 40.9
- 13 Celeris-1 Decision (Celeris) 39.5
- 14 imajev 9B 34.5
- 15 H2O-Lightning-4B (image) 34.1
- 16 Wity-1 (AlphaNimble), reasoning off off 33.1
- 17 Jev-Omni 32.8
- 18 JPT-9B (kirp / llm2jev) 31.4
- 19 NeoHorse Jev 4B 30.4
- 20 Jevify Gemma 4 26B-A4B 29.7
- 21 AutoJev-27B 29.3
- 22 JPT-4B (kirp / llm2jev) 26.6
- 23 Jev 1.13.0 + images, preview (TypeSafe) 26.4
- 24 Jev-Vision 8B (SeanLiu) 25.5
- 25 Jev 1.14.0 + images, preview (TypeSafe) 24.3
- 26 shisa-de-1 22.8
- 27 Clef-Flash (Cloudflare) 20.4
- 28 Clef (Cloudflare) 18.6
- 29 GPT-6 Luna 17.3
- 30 Winnow-12B (Q8_0 + F16 vision projector) 17.1
- 31 Image Hopper (HopitAI) 15.3
- 32 OmniJev-Qwen3.5-9B-v4 (tzcfly) 15.0
- 33 OmniJev 4B (tinnel123) 14.8
- 34 Reflex 4B (released stable configuration) 13.8
- 35 Bonsai-2-27B v2 PQ2_0 + Q8_0 MMProj 12.0
- 36 CUA-S1 4B 11.5
- 37 Visual-Jev generic (zero-shot Qwen3.5-4B baseline) 11.1
- 38 Mapika decider-2b-vision BF16 9.52
- 39 vjev-vision (yah01) 8.86
- 40 Visual-Jev 4B Answer-SFT 8.71
- 41 GPT-5.6 Luna 7.87
- 42 Jevify Gemma 4 E4B 7.26
- 43 JPT-0.8B (kirp / llm2jev) 7.17
- 44 Glance (frozen Qwen3-VL-4B) 6.85
- 45
Gemini 3.1 Flash-Lite 6.45
- 46 diffusiongemma-26b djev v10 step160 (snowicarus) 5.90
- 47 imajev 2B 5.76
- 48 Gevva E4B (text checkpoint, image input) 4.14
- 49 Liquid d1-3B (Liquid AI) 4.10
- 50 djev-distill-v4 (tarsur385) 3.62
- 51 Jevify Qwen3-VL-2B T2 3.29
- 52 OmniJev 2B (tinnel123) 3.17
- 53 Gevva E2B multimodal 2.02
- 54 jev-spatial (Fr0zencr4nE) 1.97
- 55 JEVision (divyanshx11, visual route) 1.84
- 56 OmniJev 0.8B (tinnel123) 1.57
- 57 PlayJev 0.8B 0.50
- 58 Liquid d1-omni-600M (Liquid AI) 0.24
- 59 djev-dev BF16 0.04
- 60 Qevi-2B (MeerDevelopment) 0.03
- 61 djev-spark NVFP4 0.00
- 62
Gemini 3.8 Flash 0.00
- 63 OpenJev 4B NLI v2 (official image-premise path) 0.00
- 64 OpenJev 4B NLI v5 0.00
Source: Official ImageJevBench v0.3.0 leaderboard
- 1 GPT-6 Luna 17.3
- 2 GPT-5.6 Luna 7.87
- 3
Gemini 3.1 Flash-Lite 6.45
- 4
Gemini 3.8 Flash 0.00
Source: Official ImageJevBench v0.3.0 leaderboard
| # | Entry | Score | Setting | Released | Run on | Link | Source |
|---|---|---|---|---|---|---|---|
| 1 | kettei-fat (Sqwish) closed weights | 75.4 | Oct 2026 | docs.sqwish.ai | Official leaderboard | ||
| 2 | Wity-1 (AlphaNimble), reasoning auto closed weights | 71.2 | auto | Oct 2026 | wity.alphanimble.com | Official leaderboard | |
| 3 | H2O-Lightning-4B v1.4.1 (image) | 66.6 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 4 | Imajev-4B | 63.7 | Oct 2026 | Official leaderboard | |||
| 5 | kettei-chub (Sqwish) closed weights | 63.3 | Oct 2026 | docs.sqwish.ai | Official leaderboard | ||
| 6 | H2O-Lightning-31B v1.0.7 (image) | 51.2 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 7 | H2O-Lightning-31B v1.3.4 @c35ed950 (image) | 51.0 | Nov 2026 | huggingface.co | Official leaderboard | ||
| 8 | Sage 1.4 + images, beta (Levanto Labs) closed weights | 47.9 | Oct 2026 | docs.levanto.ai | Official leaderboard | ||
| 9 | Drex Max (Nace AI) closed weights | 44.6 | Oct 2026 | drex.nace.ai | Official leaderboard | ||
| 10 | Surogate Rune 26B-A4B v3 | 42.5 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 11 | Autoloops – Gemma 4 31B IT closed weights | 41.4 | Oct 2026 | Official leaderboard | |||
| 12 | basal-omni-1.0-ultra (Remek Kinas) | 40.9 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 13 | Celeris-1 Decision (Celeris) closed weights | 39.5 | Oct 2026 | docs.celeris.ai | Official leaderboard | ||
| 14 | imajev 9B | 34.5 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 15 | H2O-Lightning-4B (image) | 34.1 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 16 | Wity-1 (AlphaNimble), reasoning off closed weights | 33.1 | off | Oct 2026 | wity.alphanimble.com | Official leaderboard | |
| 17 | Jev-Omni | 32.8 | Oct 2026 | Official leaderboard | |||
| 18 | JPT-9B (kirp / llm2jev) | 31.4 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 19 | NeoHorse Jev 4B | 30.4 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 20 | Jevify Gemma 4 26B-A4B | 29.7 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 21 | AutoJev-27B | 29.3 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 22 | JPT-4B (kirp / llm2jev) | 26.6 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 23 | Jev 1.13.0 + images, preview (TypeSafe) closed weights | 26.4 | Oct 2026 | docs.typesafe.ai | Official leaderboard | ||
| 24 | Jev-Vision 8B (SeanLiu) | 25.5 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 25 | Jev 1.14.0 + images, preview (TypeSafe) closed weights | 24.3 | Oct 2026 | docs.typesafe.ai | Official leaderboard | ||
| 26 | shisa-de-1 | 22.8 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 27 | Clef-Flash (Cloudflare) | 20.4 | Oct 2026 | Official leaderboard | |||
| 28 | Clef (Cloudflare) | 18.6 | Oct 2026 | Official leaderboard | |||
| 29 | GPT-6 Luna | 17.3 | OpenAI Decisions | Sep 2026 | Oct 2026 | Official leaderboard | |
| 30 | Winnow-12B (Q8_0 + F16 vision projector) | 17.1 | Oct 2026 | Official leaderboard | |||
| 31 | Image Hopper (HopitAI) | 15.3 | Oct 2026 | Official leaderboard | |||
| 32 | OmniJev-Qwen3.5-9B-v4 (tzcfly) | 15.0 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 33 | OmniJev 4B (tinnel123) | 14.8 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 34 | Reflex 4B (released stable configuration) | 13.8 | Oct 2026 | Official leaderboard | |||
| 35 | Bonsai-2-27B v2 PQ2_0 + Q8_0 MMProj | 12.0 | Oct 2026 | Official leaderboard | |||
| 36 | CUA-S1 4B | 11.5 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 37 | Visual-Jev generic (zero-shot Qwen3.5-4B baseline) | 11.1 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 38 | Mapika decider-2b-vision BF16 | 9.52 | Oct 2026 | Official leaderboard | |||
| 39 | vjev-vision (yah01) | 8.86 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 40 | Visual-Jev 4B Answer-SFT | 8.71 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 41 | GPT-5.6 Luna | 7.87 | Jul 2026 | Oct 2026 | Official leaderboard | ||
| 42 | Jevify Gemma 4 E4B | 7.26 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 43 | JPT-0.8B (kirp / llm2jev) | 7.17 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 44 | Glance (frozen Qwen3-VL-4B) | 6.85 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 45 | | 6.45 | Mar 2026 | Oct 2026 | Official leaderboard | ||
| 46 | diffusiongemma-26b djev v10 step160 (snowicarus) | 5.90 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 47 | imajev 2B | 5.76 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 48 | Gevva E4B (text checkpoint, image input) | 4.14 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 49 | Liquid d1-3B (Liquid AI) | 4.10 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 50 | djev-distill-v4 (tarsur385) | 3.62 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 51 | Jevify Qwen3-VL-2B T2 | 3.29 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 52 | OmniJev 2B (tinnel123) | 3.17 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 53 | Gevva E2B multimodal | 2.02 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 54 | jev-spatial (Fr0zencr4nE) | 1.97 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 55 | JEVision (divyanshx11, visual route) | 1.84 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 56 | OmniJev 0.8B (tinnel123) | 1.57 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 57 | PlayJev 0.8B | 0.50 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 58 | Liquid d1-omni-600M (Liquid AI) | 0.24 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 59 | djev-dev BF16 | 0.04 | Oct 2026 | Official leaderboard | |||
| 60 | Qevi-2B (MeerDevelopment) | 0.03 | Oct 2026 | huggingface.co | Official leaderboard | ||
| 61 | djev-spark NVFP4 | 0.00 | Oct 2026 | Official leaderboard | |||
| 62 | | 0.00 | Sep 2026 | Oct 2026 | Official leaderboard | ||
| 63 | OpenJev 4B NLI v2 (official image-premise path) | 0.00 | Oct 2026 | Official leaderboard | |||
| 64 | OpenJev 4B NLI v5 | 0.00 | Oct 2026 | huggingface.co | Official leaderboard |
| # | Model | Score | Setting | Released | Run on | Link | Source |
|---|---|---|---|---|---|---|---|
| 1 | GPT-6 Luna | 17.3 | OpenAI Decisions | Sep 2026 | Oct 2026 | Official leaderboard | |
| 2 | GPT-5.6 Luna | 7.87 | Jul 2026 | Oct 2026 | Official leaderboard | ||
| 3 | | 6.45 | Mar 2026 | Oct 2026 | Official leaderboard | ||
| 4 | | 0.00 | Sep 2026 | Oct 2026 | Official leaderboard |
Rankings by part
We read 2 parts from the benchmark’s own leaderboard: Public accuracy and Sealed accuracy. Each ranks models on one part of the test. GPT-5.6 Luna leads both. Arrows show how far a model moved from its overall rank among the models we track.
Public accuracy
4 models we track have a score here.
- 1 GPT-5.6 Luna ↑1 86.4%
- 2 GPT-6 Luna ↓1 83.4%
- 3
Gemini 3.1 Flash-Lite 58.7%
- 4
Gemini 3.8 Flash 36.6%
Sealed accuracy
4 models we track have a score here.
- 1 GPT-5.6 Luna ↑1 84.2%
- 2 GPT-6 Luna ↓1 82.1%
- 3
Gemini 3.1 Flash-Lite 67.5%
- 4
Gemini 3.8 Flash 39.7%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official ImageJevBench v0.3.0 leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official ImageJevBench v0.3.0 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.
Versions
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 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access Public tasks, plus a private test set.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (64 entries).
- Pass Models scored 4 models we track, plus 60 other entries on its leaderboard.
- Pass Newest score The newest is from September 2026, for GPT-6 Luna.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.8 Flash (#24).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on benchmarkheaven.com: 5 rows for models we track and 60 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.