Benchmarking — Kalkazu's monthly AI survey
September 2026 edition · data frozen on
Highlights of the month
First edition. Month-over-month changes start in October 2026.
Top 20 artificial intelligence models
The 20 most downloaded AI models in the world, according to Hugging Face
- 1all-MiniLM-L6-v2Empresa: sentence-transformers255,6M
- 2ms-marco-MiniLM-L6-v2Empresa: cross-encoder88,7M
- 3bge-small-en-v1.5Empresa: BAAI64,5M
- 4electra-base-discriminatorEmpresa: google56,4M
- 5bert-base-uncasedEmpresa: google-bert47,6M
- 6paraphrase-multilingual-MiniLM-L12-v2Empresa: sentence-transformers45,7M
- 7bge-m3Empresa: BAAI38,1M
- 8t5-smallEmpresa: google-t525,0M
- 9all-mpnet-base-v2Empresa: sentence-transformers23,2M
- 10chronos-2Empresa: amazon22,3M
- 11Qwen3-0.6BEmpresa: Qwen22,2M
- 12xlm-roberta-baseEmpresa: FacebookAI21,9M
- 13clip-vit-base-patch32Empresa: openai21,8M
- 14MiniMax-H3Empresa: Comfy-Org20,0M
- 15Qwen3-VL-8B-InstructEmpresa: Qwen18,8M
- 16bge-reranker-v2-m3Empresa: BAAI17,8M
- 17mobilenetv3_small_100.lamb_in1kEmpresa: timm17,5M
- 18wav2vec2-large-xlsr-53-japaneseEmpresa: jonatasgrosman17,3M
- 19gpt2Empresa: openai-community15,6M
- 20nomic-embed-text-v1.5Empresa: nomic-ai15,0M
Updated on Sep 18, 2026 · Click a model to see its full profile
Most Downloaded
| # | Model | Organization | Value |
|---|---|---|---|
| 1 | all-MiniLM-L6-v2 | sentence-transformers | 255,6M |
| 2 | ms-marco-MiniLM-L6-v2 | cross-encoder | 88,7M |
| 3 | bge-small-en-v1.5 | BAAI | 64,5M |
| 4 | electra-base-discriminator | 56,4M | |
| 5 | bert-base-uncased | google-bert | 47,6M |
| 6 | paraphrase-multilingual-MiniLM-L12-v2 | sentence-transformers | 45,7M |
| 7 | bge-m3 | BAAI | 38,1M |
| 8 | t5-small | google-t5 | 25,0M |
| 9 | all-mpnet-base-v2 | sentence-transformers | 23,2M |
| 10 | chronos-2 | amazon | 22,3M |
| 11 | Qwen3-0.6B | Qwen | 22,2M |
| 12 | xlm-roberta-base | FacebookAI | 21,9M |
| 13 | clip-vit-base-patch32 | openai | 21,8M |
| 14 | MiniMax-H3 | Comfy-Org | 20,0M |
| 15 | Qwen3-VL-8B-Instruct | Qwen | 18,8M |
| 16 | bge-reranker-v2-m3 | BAAI | 17,8M |
| 17 | mobilenetv3_small_100.lamb_in1k | timm | 17,5M |
| 18 | wav2vec2-large-xlsr-53-japanese | jonatasgrosman | 17,3M |
| 19 | gpt2 | openai-community | 15,6M |
| 20 | nomic-embed-text-v1.5 | nomic-ai | 15,0M |
Most Liked
| # | Model | Organization | Value |
|---|---|---|---|
| 1 | Qwen3.8-27B | Qwen | 15,5K |
| 2 | FLUX.1-dev | black-forest-labs | 14,8K |
| 3 | DeepSeek-R1 | deepseek-ai | 14,3K |
| 4 | Kimi-K3 | moonshotai | 11,4K |
| 5 | stable-diffusion-xl-base-1.0 | stabilityai | 8,2K |
| 6 | Llama-3.1-8B-Instruct | meta-llama | 7,7K |
| 7 | stable-diffusion-v1-4 | CompVis | 7,1K |
| 8 | Kokoro-82M | hexgrad | 6,9K |
| 9 | Meta-Llama-3-8B | meta-llama | 6,7K |
| 10 | whisper-large-v3 | openai | 6,3K |
| 11 | all-MiniLM-L6-v2 | sentence-transformers | 6,0K |
| 12 | FLUX.1-schnell | black-forest-labs | 5,9K |
| 13 | DeepSeek-V4-Pro | deepseek-ai | 5,6K |
| 14 | MiniMax-H3 | MiniMaxAI | 5,4K |
| 15 | Qwen3.8-Flash-Next | Qwen | 5,4K |
| 16 | Z-Image-Turbo | Tongyi-MAI | 5,3K |
| 17 | gpt-oss-120b | openai | 5,3K |
| 18 | GLM-5.2 | zai-org | 5,1K |
| 19 | gpt-oss-20b | openai | 5,1K |
| 20 | bloom | bigscience | 5,1K |
Latest Releases
| # | Model | Organization | Released |
|---|---|---|---|
| 1 | Pareto | Unbiased | Sep 17, 2026 |
| 2 | DeepSeek Pro | DeepSeek | Sep 14, 2026 |
| 3 | DeepSeek Flash | DeepSeek | Sep 14, 2026 |
| 4 | Schematron V2 Turbo | Inference.net | Sep 12, 2026 |
| 5 | Schematron V2 Small | Inference.net | Sep 12, 2026 |
| 6 | GPT Astra | OpenAI | Sep 11, 2026 |
| 7 | GPT Sol | OpenAI | Sep 11, 2026 |
| 8 | GPT Terra | OpenAI | Sep 11, 2026 |
| 9 | GPT Luna | OpenAI | Sep 11, 2026 |
| 10 | Fugu Ultra v2 | Sakana | Sep 11, 2026 |
| 11 | Fugu Max | Sakana | Sep 11, 2026 |
| 12 | Ling 3.0 Flash VL | inclusionAI | Sep 10, 2026 |
| 13 | DeepSeek V4.1 Flash | DeepSeek | Sep 10, 2026 |
| 14 | Mercury 2.5 | Inception | Sep 8, 2026 |
| 15 | Nex-N2.5-Mini | Nex AGI | Sep 8, 2026 |
| 16 | Nex-N2.5-Pro | Nex AGI | Sep 8, 2026 |
| 17 | GPT-6 Astra | OpenAI | Sep 4, 2026 |
| 18 | GPT-6 Astra Pro | OpenAI | Sep 4, 2026 |
| 19 | Ling 3.0 Flash Sante | inclusionAI | Sep 4, 2026 |
| 20 | Qwen3.8 Max (0902) | Qwen | Sep 3, 2026 |
How to cite
Kalkazu, Benchmarking — September 2026. kalkazu.org/en/benchmarking/2026-09
Download the data
Methodology
Most Downloaded and Most Liked are the 20 open models with the most downloads and the most likes on Hugging Face. Latest Releases are the 20 most recent AI models in the OpenRouter catalog, which includes closed models. They are different universes and do not form a single ranking.
In the releases list, variants of the same model (free, batch, fast, dates at the end of the name) are merged before the cut to 20. The month-over-month change compares each model by its address at the source, and only applies to downloads and likes.
Each edition is frozen on the 1st of every month at 03:00 Brasília time (06:00 UTC) and does not change afterwards. Actual date of the data in this edition: Most Downloaded on , Most Liked on , Latest Releases on .
This edition, the first one, was frozen on the day the survey was launched.
License CC BY 4.0: you may use and republish the data, as long as you credit Kalkazu.