BabyVision
The official leaderboard hasn’t scored a model released after Qwen 3.8 Max (August 2026). Labs report their own numbers for 11 newer models, such as Kimi K3.1 Pro at 95.4%.
Model solves 388 perception-heavy visual puzzles (discrimination, tracking, spatial, patterns) that young children solve. Accuracy versus 94% human baseline.
Moonshot AI is a listed collaborator.
- Released
- January 2026
- Kind
- Spatial reasoning and counting
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision
- Strongest model tested
- Claude Fable 5 #12 on the capability index
- Last in a launch post
- November 2026
- Last model added
- November 2026
- Scores refreshed
- November 20th, 2026
In the authors’ words
While humans develop core visual skills long before acquiring language, contemporary Multimodal LLMs (MLLMs) still rely heavily on linguistic priors to compensate for their fragile visual understanding. We uncovered a crucial fact: state-of-the-art MLLMs consistently fail on basic visual tasks that humans, even 3-year-olds, can solve effortlessly. To systematically investigate this gap, we introduce BabyVision, a benchmark designed to assess core visual abilities independent of linguistic knowledge for MLLMs. BabyVision spans a wide range of tasks, with 388 items divided into 22 subclasses across four key categories. Empirical results and human evaluation reveal that leading MLLMs perform significantly below human baselines. Gemini3-Pro-Preview scores 49.7, lagging behind 6-year-old humans and falling well behind the average adult score of 94.1. These results show despite excelling in knowledge-heavy evaluations, current MLLMs still lack fundamental visual primitives. Progress in BabyVision represents a step toward human-level visual perception and reasoning capabilities. We also explore solving visual reasoning with generation models by proposing BabyVision-Gen and automatic evaluation toolkit. Our code and benchmark data are released at https://github.com/UniPat-AI/BabyVision for reproduction.
Qwen 3.8 Max leads with 91.3% among models released by August 2026 Qwen 3.8 Max leads with 91.3% among models released by August 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. This view shows only models we track, each at its best.
Kimi K3.1 Pro leads with 95.4%, claims included
Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.
The newest model with a score here came out in August 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 Astra
- and 14 more
- 91.3
Qwen 3.8 Max
- 90.5 Claude Fable 5
- 88.9 GPT-5.6 Sol
- 85.7 Kimi K3
- 83.6 GPT-5.5
- 81.2 Claude Opus 4.8
- 76.3
Muse Spark 1.1
- 73.7
Seed2.1 Pro
- 70.4
Qwen3.7-Plus
- 68.5 Kimi K2.6
- 60.6
Seed2.0 Pro
- 51.6
Gemini 3.1 Pro
- 49.7
Gemini 3 Pro
- 49.7 GPT-5.4
- 43.3
Qwen3.5 397B-A17B
- 36.5 Kimi K2.5
- 34.4 GPT-5.2
- 30.2
Seed1.8
- 22.2
Qwen3-VL-235B-A22B
- 19.2
InternVL3.5-241B-A28B
- 19.2
Qwen3-VL-Plus
- 17.6 GLM-4.6V
- 16.2 Grok 4
- 15.1 MiMo-VL-7B-RL
- 14.8 Claude Opus 4.6
- 14.7
Step 3
- 14.2 Claude Opus 4.5
- 12.4 KimiVL-A3B
- 1
Qwen 3.8 Max 91.3%
- 2 Claude Fable 5 max 90.5%
- 3 GPT-5.6 Sol max 88.9%
- 4 Kimi K3 85.7%
- 5 GPT-5.5 xhigh 83.6%
- 6 Claude Opus 4.8 max 81.2%
- 7
Muse Spark 1.1 76.3%
- 8
Seed2.1 Pro 73.7%
- 9
Qwen3.7-Plus 70.4%
- 10 Kimi K2.6 68.5%
- 11
Seed2.0 Pro high 60.6%
- 12
Gemini 3.1 Pro 51.6%
- 13
Gemini 3 Pro 49.7%
- 14 GPT-5.4 xhigh 49.7%
- 15
Qwen3.5 397B-A17B 43.3%
- 16 Kimi K2.5 36.5%
- 17 GPT-5.2 34.4%
- 18
Seed1.8 30.2%
- 19
Qwen3-VL-235B-A22B 22.2%
- 20
InternVL3.5-241B-A28B 19.2%
- 21
Qwen3-VL-Plus 19.2%
- 22 GLM-4.6V 17.6%
- 23 Grok 4 16.2%
- 24 MiMo-VL-7B-RL 15.1%
- 25 Claude Opus 4.6 14.8%
- 26
Step 3 14.7%
- 27 Claude Opus 4.5 14.2%
- 28 KimiVL-A3B 12.4%
Source: Official BabyVision leaderboard
- 91.3
Qwen 3.8 Max
- 90.5 Claude Fable 5
- 88.9 GPT-5.6 Sol
- 85.7 Kimi K3
- 83.6 GPT-5.5
- 81.2 Claude Opus 4.8
- 76.3
Muse Spark 1.1
- 73.7
Seed2.1 Pro
- 70.4
Qwen3.7-Plus
- 68.5 Kimi K2.6
- 60.6
Seed2.0 Pro
- 51.6
Gemini 3.1 Pro
- 49.7
Gemini 3 Pro
- 49.7 GPT-5.4
- 43.3
Qwen3.5 397B-A17B
- 36.5 Kimi K2.5
- 34.4 GPT-5.2
- 30.2
Seed1.8
- 22.2
Qwen3-VL-235B-A22B
- 19.2
InternVL3.5-241B-A28B
- 19.2
Qwen3-VL-Plus
- 17.6 GLM-4.6V
- 16.2 Grok 4
- 15.1 MiMo-VL-7B-RL
- 14.8 Claude Opus 4.6
- 14.7
Step 3
- 14.2 Claude Opus 4.5
- 1
Qwen 3.8 Max 91.3%
- 2 Claude Fable 5 max 90.5%
- 3 GPT-5.6 Sol max 88.9%
- 4 Kimi K3 85.7%
- 5 GPT-5.5 xhigh 83.6%
- 6 Claude Opus 4.8 max 81.2%
- 7
Muse Spark 1.1 76.3%
- 8
Seed2.1 Pro 73.7%
- 9
Qwen3.7-Plus 70.4%
- 10 Kimi K2.6 68.5%
- 11
Seed2.0 Pro high 60.6%
- 12
Gemini 3.1 Pro 51.6%
- 13
Gemini 3 Pro 49.7%
- 14 GPT-5.4 xhigh 49.7%
- 15
Qwen3.5 397B-A17B 43.3%
- 16 Kimi K2.5 36.5%
- 17 GPT-5.2 34.4%
- 18
Seed1.8 30.2%
- 19
Qwen3-VL-235B-A22B 22.2%
- 20
InternVL3.5-241B-A28B 19.2%
- 21
Qwen3-VL-Plus 19.2%
- 22 GLM-4.6V 17.6%
- 23 Grok 4 16.2%
- 24 MiMo-VL-7B-RL 15.1%
- 25 Claude Opus 4.6 14.8%
- 26
Step 3 14.7%
- 27 Claude Opus 4.5 14.2%
Source: Official BabyVision leaderboard
- 95.4 Kimi K3.1 Pro
- 93.8
Qwen3.8 Max (0902)
- 92.5
DeepSeek V4.1 Flash 1017
- 91.3
Qwen 3.8 Max
- 91.0
MiniMax-M3.1-Flash
- 90.5 Claude Fable 5
- 89.6
DeepSeek V4.1 Flash
- 88.9 GPT-5.6 Sol
- 85.7 Kimi K3
- 85.6
Qwen 3.8 27B
- 83.6 GPT-5.5
- 81.2 Claude Opus 4.8
- 80.2
Muse Spark Open
- 76.3
Muse Spark 1.1
- 73.7
Seed2.1 Pro
- 70.4
Qwen3.7-Plus
- 68.6
Inkling 2 Small
- 68.5 Kimi K2.6
- 64.7
Seed2.0 Lite (0428)
- 60.6
Seed2.0 Pro
- 53.4 GLM-5.3-Flash
- 53.4
Inkling 2 Mini
- 52.3
Qwen3.5 397B-A17B
- 51.6
Gemini 3.1 Pro
- 49.7
Gemini 3 Pro
- 49.7 GPT-5.4
- 44.6
Qwen3.5-27B
- 40.2
Qwen3.5-122B-A10B
- 38.4
Qwen3.5-35B-A3B
- 36.5 Kimi K2.5
- 34.4 GPT-5.2
- 30.2
Seed1.8
- 29.6
Nemotron 3.5 Super VL
- 28.6
Qwen3.5-9B
- 22.2
Qwen3-VL-235B-A22B
- 19.2
InternVL3.5-241B-A28B
- 19.2
Qwen3-VL-Plus
- 18.8
EXAONE 4.5 33B
- 17.6 GLM-4.6V
- 16.2 Grok 4
- 16.0
Qwen3.5-4B
- 15.1 MiMo-VL-7B-RL
- 14.8 Claude Opus 4.6
- 14.7
Step 3
- 14.2 Claude Opus 4.5
- 1 Kimi K3.1 Pro 95.4%
- 2
Qwen3.8 Max (0902) 93.8%
- 3
DeepSeek V4.1 Flash 1017 92.5%
- 4
Qwen 3.8 Max 91.3%
- 5
MiniMax-M3.1-Flash 91%
- 6 Claude Fable 5 max 90.5%
- 7
DeepSeek V4.1 Flash 89.6%
- 8 GPT-5.6 Sol max 88.9%
- 9 Kimi K3 85.7%
- 10
Qwen 3.8 27B 85.6%
- 11 GPT-5.5 xhigh 83.6%
- 12 Claude Opus 4.8 max 81.2%
- 13
Muse Spark Open 80.2%
- 14
Muse Spark 1.1 76.3%
- 15
Seed2.1 Pro 73.7%
- 16
Qwen3.7-Plus 70.4%
- 17
Inkling 2 Small 68.6%
- 18 Kimi K2.6 68.5%
- 19
Seed2.0 Lite (0428) 64.7%
- 20
Seed2.0 Pro high 60.6%
- 21 GLM-5.3-Flash 53.4%
- 22
Inkling 2 Mini 53.4%
- 23
Qwen3.5 397B-A17B 52.3%
- 24
Gemini 3.1 Pro 51.6%
- 25
Gemini 3 Pro 49.7%
- 26 GPT-5.4 xhigh 49.7%
- 27
Qwen3.5-27B 44.6%
- 28
Qwen3.5-122B-A10B 40.2%
- 29
Qwen3.5-35B-A3B 38.4%
- 30 Kimi K2.5 36.5%
- 31 GPT-5.2 34.4%
- 32
Seed1.8 30.2%
- 33
Nemotron 3.5 Super VL 29.6%
- 34
Qwen3.5-9B 28.6%
- 35
Qwen3-VL-235B-A22B 22.2%
- 36
InternVL3.5-241B-A28B 19.2%
- 37
Qwen3-VL-Plus 19.2%
- 38
EXAONE 4.5 33B 18.8%
- 39 GLM-4.6V 17.6%
- 40 Grok 4 16.2%
- 41
Qwen3.5-4B 16%
- 42 MiMo-VL-7B-RL 15.1%
- 43 Claude Opus 4.6 14.8%
- 44
Step 3 14.7%
- 45 Claude Opus 4.5 14.2%
Sources: Labs’ model cards and official BabyVision leaderboard
| # | Entry | Score | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 91.3% | tools | Aug 2026 | Aug 2026 | Official leaderboard |
| 2 | Claude Fable 5 | 90.5% | max, tools | Jun 2026 | Jun 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 88.9% | max, tools | Jul 2026 | Jun 2026 | Official leaderboard |
| 4 | Kimi K3 | 85.7% | Python tools | Jul 2026 | Jul 2026 | Official leaderboard |
| 5 | GPT-5.5 | 83.6% | xhigh, tools | Apr 2026 | Apr 2026 | Official leaderboard |
| 6 | Claude Opus 4.8 | 81.2% | max, tools | May 2026 | May 2026 | Official leaderboard |
| 7 | | 76.3% | tools | Jul 2026 | Jul 2026 | Official leaderboard |
| 8 | | 73.7% | Jun 2026 | Jun 2026 | Official leaderboard | |
| 9 | | 70.4% | tools | Jun 2026 | Jun 2026 | Official leaderboard |
| 10 | Kimi K2.6 | 68.5% | tools | Apr 2026 | Apr 2026 | Official leaderboard |
| 11 | | 60.6% | high | Feb 2026 | Feb 2026 | Official leaderboard |
| 12 | | 51.6% | Feb 2026 | Feb 2026 | Official leaderboard | |
| 13 | | 49.7% | Nov 2025 | Nov 2025 | Official leaderboard | |
| 14 | GPT-5.4 | 49.7% | xhigh | Mar 2026 | Mar 2026 | Official leaderboard |
| 15 | | 43.3% | Feb 2026 | Feb 2026 | Official leaderboard | |
| 16 | Kimi K2.5 | 36.5% | Jan 2026 | Jan 2026 | Official leaderboard | |
| 17 | GPT-5.2 | 34.4% | Dec 2025 | Dec 2025 | Official leaderboard | |
| 18 | | 30.2% | Dec 2025 | Dec 2025 | Official leaderboard | |
| 19 | | 22.2% | Sep 2025 | Sep 2025 | Official leaderboard | |
| 20 | | 19.2% | Aug 2025 | Aug 2025 | Official leaderboard | |
| 21 | | 19.2% | Sep 2025 | Sep 2025 | Official leaderboard | |
| 22 | GLM-4.6V | 17.6% | Dec 2025 | Dec 2025 | Official leaderboard | |
| 23 | Grok 4 | 16.2% | Jul 2025 | Jul 2025 | Official leaderboard | |
| 24 | MiMo-VL-7B-RL | 15.1% | May 2025 | Apr 2025 | Official leaderboard | |
| 25 | Claude Opus 4.6 | 14.8% | Feb 2026 | Feb 2026 | Official leaderboard | |
| 26 | | 14.7% | Jul 2025 | Jul 2025 | Official leaderboard | |
| 27 | Claude Opus 4.5 | 14.2% | Nov 2025 | Nov 2025 | Official leaderboard | |
| 28 | KimiVL-A3B open weights | 12.4% | Apr 2025 | Official leaderboard |
| # | Model | Score | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 91.3% | tools | Aug 2026 | Aug 2026 | Official leaderboard |
| 2 | Claude Fable 5 | 90.5% | max, tools | Jun 2026 | Jun 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 88.9% | max, tools | Jul 2026 | Jun 2026 | Official leaderboard |
| 4 | Kimi K3 | 85.7% | Python tools | Jul 2026 | Jul 2026 | Official leaderboard |
| 5 | GPT-5.5 | 83.6% | xhigh, tools | Apr 2026 | Apr 2026 | Official leaderboard |
| 6 | Claude Opus 4.8 | 81.2% | max, tools | May 2026 | May 2026 | Official leaderboard |
| 7 | | 76.3% | tools | Jul 2026 | Jul 2026 | Official leaderboard |
| 8 | | 73.7% | Jun 2026 | Jun 2026 | Official leaderboard | |
| 9 | | 70.4% | tools | Jun 2026 | Jun 2026 | Official leaderboard |
| 10 | Kimi K2.6 | 68.5% | tools | Apr 2026 | Apr 2026 | Official leaderboard |
| 11 | | 60.6% | high | Feb 2026 | Feb 2026 | Official leaderboard |
| 12 | | 51.6% | Feb 2026 | Feb 2026 | Official leaderboard | |
| 13 | | 49.7% | Nov 2025 | Nov 2025 | Official leaderboard | |
| 14 | GPT-5.4 | 49.7% | xhigh | Mar 2026 | Mar 2026 | Official leaderboard |
| 15 | | 43.3% | Feb 2026 | Feb 2026 | Official leaderboard | |
| 16 | Kimi K2.5 | 36.5% | Jan 2026 | Jan 2026 | Official leaderboard | |
| 17 | GPT-5.2 | 34.4% | Dec 2025 | Dec 2025 | Official leaderboard | |
| 18 | | 30.2% | Dec 2025 | Dec 2025 | Official leaderboard | |
| 19 | | 22.2% | Sep 2025 | Sep 2025 | Official leaderboard | |
| 20 | | 19.2% | Aug 2025 | Aug 2025 | Official leaderboard | |
| 21 | | 19.2% | Sep 2025 | Sep 2025 | Official leaderboard | |
| 22 | GLM-4.6V | 17.6% | Dec 2025 | Dec 2025 | Official leaderboard | |
| 23 | Grok 4 | 16.2% | Jul 2025 | Jul 2025 | Official leaderboard | |
| 24 | MiMo-VL-7B-RL | 15.1% | May 2025 | Apr 2025 | Official leaderboard | |
| 25 | Claude Opus 4.6 | 14.8% | Feb 2026 | Feb 2026 | Official leaderboard | |
| 26 | | 14.7% | Jul 2025 | Jul 2025 | Official leaderboard | |
| 27 | Claude Opus 4.5 | 14.2% | Nov 2025 | Nov 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Kimi K3.1 Pro | 95.4% | Oct 2026 | Lab model card | |
| 2 | 93.8% | Sep 2026 | Lab model card | ||
| 3 | 92.5% | Oct 2026 | Lab model card | ||
| 4 | 91.3% | tools | Aug 2026 | Official leaderboard | |
| 5 | 91% | Oct 2026 | Lab model card | ||
| 6 | Claude Fable 5 | 90.5% | max, tools | Jun 2026 | Official leaderboard |
| 7 | 89.6% | Sep 2026 | Lab model card | ||
| 8 | GPT-5.6 Sol | 88.9% | max, tools | Jul 2026 | Official leaderboard |
| 9 | Kimi K3 | 85.7% | Python tools | Jul 2026 | Official leaderboard |
| 10 | 85.6% | Aug 2026 | Lab model card | ||
| 11 | GPT-5.5 | 83.6% | xhigh, tools | Apr 2026 | Official leaderboard |
| 12 | Claude Opus 4.8 | 81.2% | max, tools | May 2026 | Official leaderboard |
| 13 | 80.2% | Oct 2026 | Lab model card | ||
| 14 | 76.3% | tools | Jul 2026 | Official leaderboard | |
| 15 | 73.7% | Jun 2026 | Official leaderboard | ||
| 16 | 70.4% | tools | Jun 2026 | Official leaderboard | |
| 17 | 68.6% | Nov 2026 | Lab model card | ||
| 18 | Kimi K2.6 | 68.5% | tools | Apr 2026 | Official leaderboard |
| 19 | 64.7% | Apr 2026 | Lab model card | ||
| 20 | 60.6% | high | Feb 2026 | Official leaderboard | |
| 21 | GLM-5.3-Flash | 53.4% | Aug 2026 | Lab model card | |
| 22 | 53.4% | Nov 2026 | Lab model card | ||
| 23 | 52.3% | Feb 2026 | Lab model card | ||
| 24 | 51.6% | Feb 2026 | Official leaderboard | ||
| 25 | 49.7% | Nov 2025 | Official leaderboard | ||
| 26 | GPT-5.4 | 49.7% | xhigh | Mar 2026 | Official leaderboard |
| 27 | 44.6% | Feb 2026 | Lab model card | ||
| 28 | 40.2% | Feb 2026 | Lab model card | ||
| 29 | 38.4% | Feb 2026 | Lab model card | ||
| 30 | Kimi K2.5 | 36.5% | Jan 2026 | Official leaderboard | |
| 31 | GPT-5.2 | 34.4% | Dec 2025 | Official leaderboard | |
| 32 | 30.2% | Dec 2025 | Official leaderboard | ||
| 33 | 29.6% | Oct 2026 | Lab model card | ||
| 34 | 28.6% | Mar 2026 | Lab model card | ||
| 35 | 22.2% | Sep 2025 | Official leaderboard | ||
| 36 | 19.2% | Aug 2025 | Official leaderboard | ||
| 37 | 19.2% | Sep 2025 | Official leaderboard | ||
| 38 | 18.8% | Apr 2026 | Lab model card | ||
| 39 | GLM-4.6V | 17.6% | Dec 2025 | Official leaderboard | |
| 40 | Grok 4 | 16.2% | Jul 2025 | Official leaderboard | |
| 41 | 16% | Mar 2026 | Lab model card | ||
| 42 | MiMo-VL-7B-RL | 15.1% | May 2025 | Official leaderboard | |
| 43 | Claude Opus 4.6 | 14.8% | Feb 2026 | Official leaderboard | |
| 44 | 14.7% | Jul 2025 | Official leaderboard | ||
| 45 | Claude Opus 4.5 | 14.2% | Nov 2025 | Official leaderboard |
Labs report scores for 27 models in their cards. Switch the scores above to rank them together.
6 models have held number one on the official leaderboard since its release in January 2026 8 models have held number one with claims included since its release in January 2026
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.
Sources: Official BabyVision leaderboard and labs’ model cards
Claude Fable 5 on the official leaderboard passed 85% in June 2026
That was 5 months after its release. By the same age, 25% of the benchmarks from 2026 had passed 85%. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Sources: Official BabyVision leaderboard and labs’ model cards
9 labs put it in a launch post in the last six months
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Nov 5, 2026 | 68.6 | BabyVision (original; 93.7 with Python) | |
| Oct 24, 2026 | Kimi K3.1 Pro | 95.4 | BabyVision w/ python |
| Oct 19, 2026 | 91 | BabyVision (with tools) | |
| Oct 17, 2026 | 92.5 | BabyVision (w/tools) | |
| Oct 15, 2026 | 80.2 | BabyVision (w/ tools) | |
| Oct 15, 2026 | 29.6 | BabyVision | |
| Sep 10, 2026 | 89.6 | BabyVision w/ tools (Pass@1) | |
| Sep 2, 2026 | 93.8 | General visual reasoning, BabyVision (w/ CI) | |
| Aug 26, 2026 | GLM-5.3-Flash | 53.4 | BabyVision |
| Aug 14, 2026 | 85.6 | General visual reasoning BabyVision Without CI 65.7 With CI 85.6 | |
| Aug 3, 2026 | 91.3 | BabyVision (without CI / with CI) 82.0 / 91.3 | |
| Jul 16, 2026 | Kimi K3 | 85.7 | BabyVision w/ python |
| Jul 9, 2026 | 76.3 | BabyVision (Visual reasoning) | |
| Jun 23, 2026 | 73.7 | BabyVision | |
| Jun 1, 2026 | 70.4 | BabyVision 70.4 / 64.7 (with CI / without CI) | |
| Apr 28, 2026 | 64.7 | BabyVision | |
| Apr 20, 2026 | Kimi K2.6 | 39.8 | BabyVision |
| Apr 9, 2026 | 18.8 | BabyVision | |
| Mar 2, 2026 | 16 | BabyVision 16.0/19.1 (with CI / without CI) | |
| Mar 2, 2026 | 28.6 | BabyVision 28.6/25.8 (with CI / without CI) | |
| Feb 24, 2026 | 38.4 | BabyVision 38.4 / 29.6 (with CI / without CI) | |
| Feb 24, 2026 | 44.6 | BabyVision 44.6 / 34.8 (with CI / without CI) | |
| Feb 24, 2026 | 40.2 | BabyVision 40.2 / 34.5 (with CI / without CI) | |
| Feb 16, 2026 | 52.3 | BabyVision (52.3/43.3) | |
| Feb 14, 2026 | 60.6 | BabyVision |
9 system cards report it, from 7 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 20, 2026 | | 53.4 | BabyVision (original, 88.8 with Python) | Capabilities |
| Oct 15, 2026 | | 29.6 | BabyVision (Table 10) | Capabilities |
| Sep 10, 2026 | | 89.6 | BabyVision w/ tools (Pass@1) (Table 3) | Capabilities |
| Jul 27, 2026 | Kimi K3 technical report | 85.7 | BabyVision w/ Python, the only setting reported (Table 2) | Capabilities |
| Jul 9, 2026 | | 76.3 | BabyVision (w/ tools) (Figure 44) | Capabilities |
| Jun 23, 2026 | | 73.7 | BabyVision (Table 9) | Capabilities |
| Apr 9, 2026 | | 18.8 | BabyVision (Table 2, Reasoning mode) | Capabilities |
| Feb 14, 2026 | | 60.6 | BabyVision (Table 8) | Capabilities |
| Jan 30, 2026 | Kimi K2.5 technical report | 36.5 | BabyVision (Table 4) | Capabilities |
The lab's number comes out ahead in 1 of 2 pairs that differ
Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 7 of them match, which usually means one side copied the other. Every benchmark.
| Model | Lab says | Board says | Gap | What differs |
|---|---|---|---|---|
| Kimi K2.6 | 39.8% BabyVision | 68.5% Leaderboard, tools | −29 pts | The lab score excludes Python tools, while the board reports the Python-enabled run. |
| | 52.3% BabyVision (52.3/43.3) | 43.3% Leaderboard | +9.0 pts | |
| | 60.6% BabyVision | 60.6% Leaderboard, high | Same | |
| | 70.4% BabyVision 70.4 / 64.7 (with CI / without CI) | 70.4% Leaderboard, tools | Same | |
| | 73.7% BabyVision | 73.7% Leaderboard | Same | |
| | 76.3% BabyVision (Visual reasoning) | 76.3% Leaderboard, tools | Same | |
| Kimi K3 | 85.7% BabyVision w/ python | 85.7% Leaderboard, Python tools | Same | |
| | 91.3% BabyVision (without CI / with CI) 82.0 / 91.3 | 91.3% Leaderboard, tools | Same | |
| Kimi K2.5 | 36.5% BabyVision (Table 4) | 36.5% Leaderboard | Same |
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 All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (28 entries), launch posts (25 scores) and system cards (9 scores).
- Pass Models scored 45 models we track, plus 1 other entry on its leaderboard.
- Pass Newest score The newest is from November 2026, for Inkling 2 Mini.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Claude Fable 5 (#12).
- Note Headroom Saturated. The best is 95.4%, past 85% since June 2026.
Leaderboard import 3 pass
- Pass Importer We read the board on unipat.ai: 28 rows for models we track and 1 other entry 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 1 large gap of 9 pairs, each explained in our notes.
- Pass Known issues No published audit found problems.