Citation and evidence

List of open-weight models

53 min full readUpdated 5 references

This article's verification

Report a problem with this article

Edit article
More

Use this article

Raw MarkdownExplore connections

Improve this page

Revision historyDiscuss this article

Topics

AI ModelsOpen Source AI

Cite this article

This is a list of open-weight models: artificial intelligence models whose trained parameters (weights) the developer has published for download, so that the model can be run, studied and in most cases fine-tuned outside the developer's own service. It covers the 408 models and model families that have an AI Wiki article recording downloadable weights as of November 17, 2026: 374 whose weights are available in full and 34 families in which only some members are, marked "part of the family" below. Models that are offered only through a hosted service are not listed, and the list is not exhaustive: it is limited to models with their own article.

"Open weights" describes what can be downloaded, not the terms of use. The licenses in this list range from standard permissive software licenses to research-only terms, so each row gives the license of the weights and whether it allows commercial use. A license on the weights does not by itself make a model open source in the sense of the Open Source Initiative's Open Source AI Definition, which asks for the source code used to train and run the system and detailed information about the training data in addition to the parameters.[1]

How to read the list

Each row summarizes the fact box of the linked article, where the license file, model card or announcement behind it is cited. The columns mean the following.

  • Released is the release date given in the linked article; for a family it is usually the date of the first member. For some 2026 models it is the date of the hosted launch, and the weights followed days or weeks later; the linked article gives both dates.
  • Parameters is the size stated by the developer, as recorded in the linked article. Where a model activates only part of its parameters for each token, both the total and the active count are given when the developer states them.
  • Weights license is the license of the downloadable weights, which can differ from the license of the accompanying code. For a family it is normally the license of the most recent open release, and for a release with several sizes the license that covers most of them; the linked article notes the exceptions.
  • Commercial use is "Yes" when the license allows commercial use with no conditions beyond notices, attribution or share-alike terms; "With conditions" when commercial use is allowed subject to conditions such as revenue or user thresholds, an acceptable use policy, territorial exclusions or naming and display requirements; "No" when the license covers research or non-commercial use only; and "Not stated" when the repository carries no license statement for the weights.

Within each section the newest releases come first.

Licenses at a glance

The table counts the 408 entries by their Commercial use value. A partly open family is counted under the value of its open members.

Commercial useEntriesTypical licenses
Yes238Standard licenses such as Apache 2.0, MIT and CC BY 4.0, which allow commercial use, modification and redistribution subject to notice or attribution terms [2][3][4]; also the BSD licenses and OpenMDW
With conditions113Licenses written by the developer, such as the Llama Community License, the NVIDIA Open Model License and model licenses with revenue or user thresholds, and the OpenRAIL licenses, which restrict certain uses; see open-weight license comparison
No55Terms such as CC BY-NC 4.0, which does not license commercial use [5], and developers' research licenses
Not stated2Repositories with no license statement for the weights

Expanded article table

Of the 408 entries, 34 are partly open families: some of their models have downloadable weights, and the others are hosted only or were never released. The most common licenses are Apache 2.0 (139), MIT (71), CC BY-NC 4.0 (26), Llama Community License (15), OpenMDW 1.1 (11), NVIDIA Open Model License (11), Gemma Terms of Use (10), CC BY 4.0 (10).

Language models

2026

ModelDeveloperReleasedParametersWeights licenseCommercial use
ERNIE 5.5 Flash BaseBaiduNovember 5, 2026280B total, 14B activeERNIE 5.5 Flash Base Model License Agreement v1.0With conditions
LongCat-2.5Meituan LongCat teamNovember 2, 20261.6T total, about 48B activeMITYes
Ripple-65B PreviewRicursive IntelligenceNovember 2026About 65BApache 2.0Yes
LFM3Liquid AIOctober 27, 20261.73B to 2.17BLFM Open License v1.0With conditions
Kimi K3.1 ProMoonshot AIOctober 23, 20262.8T total, 104B activeKimi K3.1-Pro LicenseWith conditions
Nemotron 3.5 Bolt E4BNVIDIAOctober 23, 20268B total (4B effective)OpenMDW 1.1Yes
Recursive FlintRecursive SuperintelligenceOctober 21, 202639.8B total; approximately 2.5B active per tokenApache 2.0Yes
Spark-X2.6-4BSparkLLM / XHTokenOctober 21, 20264,112,079,360Apache 2.0Yes
Ornith-1.5-122BOrnith (ornith-ai)October 20, 2026125.1B in checkpoint metadata (named 122B)MITYes
MiniMax M3.1 FlashMiniMaxOctober 19, 2026About 428B total, 23B active per tokenMiniMax Community LicenseWith conditions
Beam (Reflection)Reflection AIOctober 16, 2026501.3B total, 23.4B activeApache 2.0Yes
Muse Spark OpenMeta Superintelligence LabsOctober 15, 2026816B total, 42B activeMuse Model LicenseWith conditions
Nemotron 3.5 BoltNVIDIAOctober 15, 202674B total, 9B activeOpenMDW 1.1Yes
Nemotron 3.5 Super VLNVIDIAOctober 15, 2026120B total, 12B active per tokenOpenMDW 1.1Yes
Underdog Saluki 27BUnderdog (Conway Research)October 7, 202627B (2-bit quantized; 7.89 GB GGUF file)CC BY-NC 4.0No
EmbeddingGemma 2Google DeepMindOctober 6, 2026About 740M (all modality encoders loaded)Apache 2.0Yes
Mistral Large 4Mistral AIOctober 6, 2026About 1T total, 49B active per tokenMistral Ecosystem Open-Weight LicenseWith conditions
Kolibri 1Aleph AlphaOctober 3, 202678.1B total, 3.46B active (excluding embeddings)Apache 2.0Yes
Alice AI Foundation 80B-A3BYandexSeptember 21, 202680B total, 3B activeApache 2.0Yes
Xiaomi MiMo-V2.6Xiaomi (MiMo team)September 21, 2026310B total, 15B active (Flash)MITYes
Step 5 PreviewStepFunSeptember 20, 2026600B total, 27B activeStepFun Model License 1.0With conditions
DeepSeek V4.1-FlashDeepSeekSeptember 10, 2026Not statedMITYes
MiniCPM5-2BOpenBMBSeptember 7, 20262.5B (2,516,756,480 total; 1,981,982,720 non-embedding)Apache 2.0Yes
Hy4 PreviewTencent Hy TeamAugust 28, 2026770B total, 49B active (backbone)Apache 2.0Yes
GLM-5.3-FlashZ.aiAugust 26, 2026320B total, 18B activeMITYes
Qwen3.8-Flash-NextQwen team, AlibabaAugust 26, 2026125B main model, 6B activeQwen Community License 1.0With conditions
GLM-5.3Z.aiAugust 14, 2026744B total, 40B active (Z.ai model-index label)GLM-5.3 LicenseWith conditions
dots3-note Previewdots studio (Xiaohongshu)August 13, 2026280B total, 16B active (language backbone)Apache 2.0Yes
Nemotron 3.5 LightningNVIDIAAugust 11, 202630B total, 3B active (31.6B / 3.6B counting embeddings)OpenMDW 1.1Yes
Muse GlimmerMeta AIAugust 10, 2026About 29.6B (29,776,626,688 in the BF16 checkpoint)Apache 2.0Yes
Qwen3.8-MaxQwen team, AlibabaAugust 3, 20262.4T total, 95B activeQwen3.8-Max LicenseWith conditions
BigBang-v1The Endless Frontier teamAugust 2, 202635B total, about 3B activeApache 2.0Yes
Ling-3.0-flashinclusionAIJuly 23, 2026124B total, 5.1B active per tokenMITYes
Qwen3.8Qwen team, Alibaba CloudJuly 19, 202627.8B (Qwen3.8-27B); 2.4T total, 95B active (Qwen3.8-Max)Qwen3.8-Max LicenseWith conditions
Kimi K3Moonshot AIJuly 16, 20262.78T total; 104.2B activated per tokenKimi K3 LicenseWith conditions
Inkling (Thinking Machines Lab)Thinking Machines LabJuly 15, 2026975B total, 41B activeApache 2.0Yes
Leanstral 1.5Mistral AIJune 30, 2026119B total, 6.5B activeApache 2.0Yes
Kimi K2.7-CodeMoonshot AIJune 12, 20261T total, 32B activeModified MITWith conditions
DiffusionGemmaGoogle DeepMindJune 10, 202625.2B total, 3.8B activeApache 2.0Yes
North Mini CodeCohereJune 9, 202630B total, about 3B activeApache 2.0Yes
MiniMax M3MiniMaxJune 1, 2026Not statedMiniMax Community LicenseWith conditions
GLM-5.2Zhipu AI (Z.ai)June 2026744B total, 40B active per Z.ai's model indexMITYes
Hy MemoryTencent Hunyuan (Tencent Hy)May 28, 2026About 3B active (Hy-Memory-30B-A3B)Apache 2.0Yes
ZAYA1-8BZyphraMay 6, 20268.4B total; 760M active per tokenApache 2.0Yes
Mistral Medium 3.5Mistral AIApril 28, 2026~128 billion (all active per token)Modified MITWith conditions
DeepSeek V4DeepSeekApril 24, 20261.6T total, 49B active (Pro); 284B total, 13B active (Flash)MITYes
DeepSeek V4-FlashDeepSeekApril 24, 2026284B total, 13B activeMITYes
DeepSeek V4-ProDeepSeekApril 24, 20261.6T total, 49B activeMITYes
Tencent Hunyuan Hy3TencentApril 23, 2026~295B total; ~21B active per tokenApache 2.0Yes
Kimi K2.6Moonshot AIApril 20, 2026~1T total; ~32B active per tokenModified MITWith conditions
Muse Spark (part of the family)Meta Superintelligence LabsApril 8, 2026816B total, 42B active (Muse Spark Open)Muse Model LicenseWith conditions
GLM-5.1Zhipu AI (Z.ai)April 7, 2026744B total, 40B active per Z.ai's model indexMITYes
Gemma 4Google DeepMindApril 2, 2026E2B 5.1B; E4B 8B; 26B A4B 25.2B (3.8B active); 31B 30.7BApache 2.0Yes
Qwen3.6 (part of the family)Alibaba (Qwen team)April 2026Open: 35B/3B active (MoE), 27B (dense); hosted undisclosedApache 2.0Yes
MiniMax M2.7MiniMaxMarch 18, 2026~230B total; ~10B active per tokenMiniMax M2.7 Non-Commercial LicenseNo
Mistral Small 4Mistral AIMarch 16, 2026119 billion total; ~6.5 billion active per tokenApache 2.0Yes
Qwen3.5Qwen team, AlibabaFebruary 16, 2026397B totalApache 2.0Yes
GLM-5Zhipu AI (Z.ai)February 11, 2026744B total; ~40B active per tokenMITYes
Qwen3-Coder-NextAlibaba (Qwen team)February 3, 2026About 80B total, about 3B active per tokenApache 2.0Yes
Kimi K2.5Moonshot AIJanuary 27, 20261T total; 32B active per tokenModified MITWith conditions
Jamba2AI21 LabsJanuary 8, 20263B dense; Mini 52B total, 12B activeApache 2.0Yes

Expanded article table

2025

ModelDeveloperReleasedParametersWeights licenseCommercial use
Nemotron 3NVIDIADecember 15, 2025Nano 30B (3.5B active); Super 120.6B (12.7B)NVIDIA Open Model LicenseWith conditions
Mistral Large 3Mistral AIDecember 2, 2025~675B total; ~41B active per token (vision encoder ~2.5B)Apache 2.0Yes
OLMo 3Allen Institute for AI (Ai2)November 20, 20257B and 32BApache 2.0Yes
Kimi K2 ThinkingMoonshot AINovember 6, 2025About 1 trillion total; about 32 billion active per tokenModified MITWith conditions
Kimi LinearMoonshot AIOctober 30, 202548B total, 3B activeMITYes
MiniMax M2MiniMaxOctober 27, 2025230B total; ~10B active per tokenModified MITWith conditions
Ling-1TAnt Group (Inclusion AI)October 9, 2025~1 trillion total; 51B active per tokenMITYes
Jamba Reasoning 3BAI21 LabsOctober 8, 2025~3BApache 2.0Yes
IBM Granite 4.0IBMOctober 2, 2025~3B (Micro) to ~32B total with ~9B active (H-Small)Apache 2.0Yes
Ring-1TinclusionAI (Ant Group)October 2025~1T total; ~50B active per tokenMITYes
GLM-4.6Zhipu AI (Z.ai)September 30, 2025~357B total; ~32B active per tokenMITYes
DeepSeek V3.2DeepSeekSeptember 29, 2025671B total; 37B active per tokenMITYes
Qwen3-VLQwen team, Alibaba CloudSeptember 23, 2025Dense 2B, 4B, 8B, 32B; MoE 30B-A3B and 235B-A22BApache 2.0Yes
EmbeddingGemmaGoogle DeepMindSeptember 4, 2025~308M (~100M backbone, ~200M embedding table)Gemma Terms of UseWith conditions
ApertusSwiss AI Initiative (EPFL, ETH Zurich, CSCS)September 2, 20258B and 70B (70B reported as 71B); base and Instruct variantsApache 2.0Yes
LongCat-FlashMeituan (LongCat team)September 1, 2025~560B total; 18.6B-31.3B active per token (avg ~27B)MITYes
Qwen3-NextQwen team, Alibaba CloudSeptember 202580 billion total; ~3 billion active per tokenApache 2.0Yes
Qwen3-OmniQwen team, Alibaba CloudSeptember 202530B total (Thinker); ~3B active per tokenApache 2.0Yes
Command A ReasoningCohere and Cohere LabsAugust 21, 2025111 billionCC BY-NC 4.0No
Jet-NemotronNVIDIA ResearchAugust 21, 2025~2B and ~4B variantsNVIDIA License (non-commercial)No
DeepSeek V3.1DeepSeekAugust 19, 2025671B total, 37B active per tokenMITYes
Nemotron Nano 2NVIDIAAugust 18, 2025~9B (flagship, pruned and distilled from a 12B base)NVIDIA Open Model LicenseWith conditions
gpt-ossOpenAIAugust 5, 2025120b: 117B total, 5.1B active; 20b: 21B total, 3.6B activeApache 2.0Yes
Hermes 4Nous ResearchAugust 202514B, 70B, 405BLlama Community LicenseWith conditions
FoxBrain (part of the family)Hon Hai Research Institute (Foxconn)July 31, 202570B (v1.x)FoxBrain Model Usage and License AgreementNo
Step-3StepFunJuly 31, 2025~321B total; ~38B active per tokenApache 2.0Yes
GLM-4.5Zhipu AI (Z.ai)July 28, 2025355B total, 32B active; Air: 106B total, 12B activeMITYes
Qwen3-CoderAlibaba Qwen team (Tongyi Lab)July 22, 2025480B total; 35B active per token (flagship)Apache 2.0Yes
Kimi K2Moonshot AIJuly 11, 20251.04T total; 32.6B active per tokenModified MITWith conditions
SmolLM 3Hugging FaceJuly 8, 20253BApache 2.0Yes
Phi-4-mini-flash-reasoningMicrosoftJuly 20253.8 billionMITYes
MiniMax M1MiniMaxJune 16, 2025456B total; 45.9B active per tokenApache 2.0Yes
Magistral (part of the family)Mistral AIJune 10, 202524B (Magistral Small); unpublished (Magistral Medium)Apache 2.0Yes
Hunyuan-A13BTencent (Hunyuan team)June 2025~80B total; ~13B active per tokenTencent Hunyuan Community LicenseWith conditions
Qwen3 EmbeddingAlibaba Qwen teamJune 20250.6B, 4B, 8BApache 2.0Yes
Devstral (part of the family)Mistral AI with All Hands AIMay 21, 202524 billion (Small); Medium not disclosedApache 2.0Yes
Falcon-H1Technology Innovation Institute (TII), Abu DhabiMay 20, 2025Six sizes: 0.5B, 1.5B, 1.5B-Deep, 3B, 7B, 34BTII Falcon LicenseWith conditions
Gemma 3nGoogle DeepMindMay 20, 20255B (E2B) and 8B (E4B) raw; about 2B and 4B effectiveGemma Terms of UseWith conditions
Phi-4 ReasoningMicrosoftApril 30, 202514 billionMITYes
Qwen3Qwen Team, Alibaba CloudApril 29, 20250.6B to 235B (flagship 235B total; 22B active)Apache 2.0Yes
InternVL3OpenGVLab (Shanghai AI Laboratory)April 11, 2025Seven dense sizes, 938M to 78.41BQwen License (Tongyi Qianwen)With conditions
Llama 4MetaApril 5, 2025Scout: 109B total, 17B activeLlama Community LicenseWith conditions
Llama 4 Scout and MaverickMetaApril 5, 2025Scout: ~109B total, 17B active (16 experts)Llama Community LicenseWith conditions
Apriel (ServiceNow)ServiceNow (AI Research / SLAM lab)April 2025~4.8B to ~15BMITYes
Nemotron-HNVIDIAApril 20258B, 47B, and 56B (base models)NVIDIA Internal Scientific R&D Model LicenseNo
Llama NemotronNvidiaMarch 18, 20258B (Nano); 49B (Super); 253B (Ultra)NVIDIA Open Model LicenseWith conditions
RWKV-7 (Goose)RWKV Project (Bo Peng et al., with EleutherAI and Recursal AI)March 18, 20250.19B to 2.9B (World v3 series); later 7.2B G1 checkpointApache 2.0Yes
ERNIE 4.5BaiduMarch 16, 20250.3B dense up to 424B total (47B active) across ten variantsApache 2.0Yes
Cohere Command ACohereMarch 13, 2025111BCC BY-NC 4.0No
Gemma 3Google DeepMindMarch 12, 20251B, 4B, 12B, 27B (270M added August 2025)Gemma Terms of UseWith conditions
Skywork-R1VSkywork AI (Kunlun Tech)March 202538B (each open generation)MITYes
Phi-4-miniMicrosoftFebruary 26, 20253.8 billionMITYes
LLaDA (Large Language Diffusion)Renmin University of China and Ant GroupFebruary 14, 20258B (LLaDA 8B); up to 100B (LLaDA 2.0 flash)MITYes
Goedel-ProverPrinceton Language and Intelligence, with Tsinghua, Amazon, Meta FAIR, NuminaFebruary 20257B (V1); 8B and 32B (V2)Apache 2.0Yes
olmOCRAllen Institute for AI (Ai2)February 20257BApache 2.0Yes
DeepSeek-R1DeepSeekJanuary 20, 2025671B total; 37B active per tokenMITYes
DeepSeek-R1-DistillDeepSeekJanuary 20, 20251.5B, 7B, 8B, 14B, 32B, and 70B variantsMITYes
MiniMax-Text-01MiniMaxJanuary 14, 2025456B total; 45.9B active per tokenMiniMax Model LicenseWith conditions

Expanded article table

2024

ModelDeveloperReleasedParametersWeights licenseCommercial use
DeepSeek V3DeepSeekDecember 26, 2024671B total (about 685B with MTP module)DeepSeek Model LicenseWith conditions
QvQ (part of the family)Qwen team, AlibabaDecember 25, 202472B (QVQ-72B-Preview)Qwen License (Tongyi Qianwen)With conditions
Falcon 3Technology Innovation Institute (TII)December 17, 20241B, 3B, 7B, 10B (transformer); 7B (Mamba)TII Falcon LicenseWith conditions
Byte Latent TransformerMeta AI (FAIR)December 13, 2024Up to 8B trained; 1B and 7B checkpoints releasedFAIR Noncommercial Research LicenseNo
Phi-4Microsoft ResearchDecember 12, 202414 billion (about 14.7B with embeddings)MITYes
Llama 3.3MetaDecember 6, 202470 billionLlama Community LicenseWith conditions
QwQAlibaba Cloud Qwen teamNovember 28, 202432.5B (dense)Apache 2.0Yes
OLMo 2Allen Institute for AI (Ai2)November 26, 20247B, 13B, 32BApache 2.0Yes
Pixtral LargeMistral AINovember 18, 2024~124 billion totalMistral Research LicenseNo
SmolLM 2Hugging FaceNovember 1, 2024135M, 360M, 1.7B (three sizes)Apache 2.0Yes
Marco-o1Alibaba International Digital Commerce (MarcoPolo team)November 2024Not statedApache 2.0Yes
Ministral 3B / 8B (part of the family)Mistral AIOctober 16, 2024~3B and ~8B (8,019,808,256 for the 8B)Apache 2.0Yes
Llama-3.1-Nemotron-70B-InstructNVIDIAOctober 202470BNVIDIA Open Model LicenseWith conditions
Llama 3.2MetaSeptember 25, 20241.23B, 3.21B (text-only); 11B, 90B (vision)Llama Community LicenseWith conditions
Llama 3.2 VisionMeta AISeptember 25, 202411B (10.6B actual) and 90B (88.8B actual)Llama Community LicenseWith conditions
Qwen2.5Alibaba Cloud Qwen teamSeptember 19, 2024Seven dense sizes from 0.49B to 72.7BApache 2.0Yes
Qwen2.5-MathQwen team, Alibaba CloudSeptember 19, 20241.5B, 7B and 72BApache 2.0Yes
Jina Embeddings v3Jina AISeptember 18, 2024570 millionCC BY-NC 4.0No
NVLMNVIDIASeptember 17, 2024~79B (NVLM-D-72B: 72B LLM plus 6B vision encoder)CC BY-NC 4.0No
PixtralMistral AISeptember 11, 2024~12.4B (12B: 12B decoder + 400M encoder); 124B (Large)Apache 2.0Yes
OLMoEAllen Institute for AI, with Contextual AISeptember 3, 20246.9B total; ~1.3B active per tokenApache 2.0Yes
Qwen2.5-CoderQwen team, AlibabaSeptember 2024Six sizes, 0.5B to 32B (flagship 32.5B)Apache 2.0Yes
Qwen2-MathQwen team, AlibabaAugust 8, 20241.5B, 7B, and 72B (base and Instruct)Apache 2.0Yes
Llama 3.1MetaJuly 23, 20248B, 70B, and 405BLlama Community LicenseWith conditions
Mistral NeMoMistral AI, with NVIDIAJuly 18, 202412BApache 2.0Yes
SmolLMHugging FaceJuly 16, 2024135M, 360M, and 1.7B variantsApache 2.0Yes
Gemma 2Google DeepMindJune 27, 20242.6B, 9B, 27BGemma Terms of UseWith conditions
LLM Compiler (Meta)Meta AIJune 27, 20247B and 13B (base and FTD variants)Meta LLM Compiler License AgreementWith conditions
Qwen2Alibaba CloudJune 6, 20240.5B, 1.5B, 7B, 57B-A14B (14B active), 72BApache 2.0Yes
Codestral (part of the family)Mistral AIMay 29, 202422.2B (22B v0.1); 7.3B (Mamba); later versions not disclosedMistral Non-Production LicenseNo
DeepSeek-ProverDeepSeekMay 23, 20247B; 671B MoE (V2)DeepSeek Model LicenseWith conditions
DeepSeek-V2DeepSeekMay 2024236B total; ~21B active per tokenDeepSeek Model LicenseWith conditions
IBM GraniteIBMMay 2024~350M to 32B MoE (H-Small, ~9B active)Apache 2.0Yes
Snowflake ArcticSnowflake AI ResearchApril 24, 2024480B total; 17B active per tokenApache 2.0Yes
Phi-3MicrosoftApril 23, 20243.8B (Mini) to 14B (Medium)MITYes
Llama 3MetaApril 18, 20248B and 70B (base and Instruct variants)Llama Community LicenseWith conditions
Mixtral 8x22BMistral AIApril 17, 2024~141B total; ~39B active per tokenApache 2.0Yes
RecurrentGemmaGoogle DeepMindApril 11, 20242.68B and 8.58B totalGemma Terms of UseWith conditions
CodeGemmaGoogleApril 9, 20242B to 7BGemma Terms of UseWith conditions
Command R+CohereApril 4, 2024104 billionCC BY-NC 4.0No
Reka Edge (part of the family)Reka AIApril 20247 billionReka Edge 2603 License (Business Source License 1.1)With conditions
JambaAI21 LabsMarch 28, 202452B total; 12B active (Jamba 1.0)Apache 2.0Yes
DBRXDatabricks (Mosaic AI)March 27, 2024132B total; 36B active per tokenDatabricks Open Model LicenseWith conditions
Command RCohereMarch 11, 202435B (Command R); family spans 7B to 104BCC BY-NC 4.0No
BitNet b1.58Microsoft ResearchFebruary 27, 2024700M to 3B (original); 2B (BitNet b1.58 2B4T)MITYes
Mistral Large (part of the family)Mistral AIFebruary 26, 2024123B (Large 2/2.1); 675B total, 41B active (Large 3)Mistral Ecosystem Open-Weight LicenseWith conditions
Nemotron-4 (part of the family)NVIDIAFebruary 26, 202415B; 340B (Base, Instruct, and Reward checkpoints)NVIDIA Open Model LicenseWith conditions
GemmaGoogle DeepMindFebruary 21, 2024270 million to 31 billion across generationsApache 2.0Yes
DeepSeekMathDeepSeekFebruary 5, 20247 billionDeepSeek Model LicenseWith conditions
MiniCPMOpenBMBFebruary 1, 20240.43B (MiniCPM4-0.5B) to 13.6BApache 2.0Yes
OLMoAllen Institute for AI (Ai2)February 1, 20241B, 7B, 13B, 32B dense; OLMoE 1B active / 7B totalApache 2.0Yes
Aya (Expanse / Vision)Cohere Labs (Cohere)February 20248B and 32B (Aya Expanse); 13B (Aya 101)CC BY-NC 4.0No
Reka Flash (part of the family)Reka AIFebruary 202421 billionApache 2.0Yes
GLM-4 (part of the family)Zhipu AIJanuary 16, 2024Not statedMITYes
MinitronNVIDIA2024About 4B to 8BNVIDIA Open Model LicenseWith conditions

Expanded article table

2023 and earlier

ModelDeveloperReleasedParametersWeights licenseCommercial use
MixtralMistral AIDecember 11, 20238x7B: 46.7B total, 12.9B activeApache 2.0Yes
MambaAlbert Gu and Tri DaoDecember 1, 2023130M to 2.8B (pretrained language models)Apache 2.0Yes
DeepSeek LLMDeepSeekNovember 29, 20237B and 67BDeepSeek Model LicenseWith conditions
Grok (part of the family)xAINovember 4, 2023314B, ~25% active (Grok-1); ~2.7T estimated (Grok-3)Grok 2 Community LicenseWith conditions
DeepSeek-CoderDeepSeekNovember 2, 20231.3B, 6.7B, 33B dense (V1)DeepSeek Model LicenseWith conditions
Yi (language model)01.AINovember 2, 20236B and 34B base (family spans 1.5B to 34B)Apache 2.0Yes
NemotronNVIDIANovember 20238B (2023) to ~550B total MoE (Nemotron 3 Ultra, 2026)OpenMDW 1.1Yes
BitNetMicrosoft ResearchOctober 17, 2023Not statedMITYes
SEDD (Score Entropy Discrete Diffusion)Stanford University (Aaron Lou, Chenlin Meng, Stefano Ermon)October 2023Not statedLicense not statedNot stated
Mistral 7BMistral AISeptember 27, 2023~7.24 billion (announced as 7.3B)Apache 2.0Yes
Hunyuan (part of the family)TencentSeptember 7, 2023389B total, 52B active (Hunyuan-Large)Apache 2.0Yes
Code LlamaMeta AIAugust 24, 20237B, 13B, 34B, 70B (base, Python, and Instruct variants)Llama Community LicenseWith conditions
Llama 2MetaJuly 18, 20237B, 13B, 70B (34B trained but withheld)Llama Community LicenseWith conditions
InternLMShanghai AI LaboratoryJuly 6, 20231.8B to 20B across generationsApache 2.0Yes
Aquila (language model)Beijing Academy of Artificial Intelligence (BAAI)June 20237B, 34B, and experimental 70BBAAI Aquila Model License AgreementWith conditions
Phi (language model)Microsoft ResearchJune 20231.3B (Phi-1) to 15B (Phi-4-reasoning-vision)MITYes
StarCoderBigCode project (Hugging Face and ServiceNow)May 4, 202315.5B (StarCoder); 3B, 7B, 15B (StarCoder2)OpenRAILWith conditions
Falcon (language model)Technology Innovation Institute (TII)May 202390M to 180B across generationsTII Falcon LicenseWith conditions
WizardLMMicrosoftApril 24, 2023Not statedLlama Community LicenseWith conditions
Qwen (part of the family)Alibaba Cloud (Qwen Team)April 11, 2023Roughly 0.5B to over 1T across variantsQwen Community License 1.0With conditions
SenseNova (part of the family)SenseTimeApril 10, 20232B to 27B (open-weight models); over 600B (SenseNova V6)Apache 2.0Yes
Vicuna (language model)LMSYSMarch 30, 20237B, 13B, 33BLlama Community LicenseWith conditions
ChatGLMZhipu AI with Tsinghua University KEG labMarch 14, 2023~6.2B (ChatGLM-6B); ~6B per later generationChatGLM Model LicenseWith conditions
CodeGeeXTsinghua University (KEG) and Zhipu AIMarch 20236B to 13BCodeGeeX4 LicenseWith conditions
LLaMAMeta AIFebruary 24, 2023Up to 65B (LLaMA 1)Llama Community LicenseWith conditions
SantaCoderBigCode (Hugging Face and ServiceNow Research)January 9, 20231.1BOpenRAILWith conditions
Atlas (language model)Meta AI (FAIR)August 2022Reader 220M to 11B, plus ~110M retrieverCC BY-NC 4.0No
GLM-130BTsinghua KEG and Zhipu AIAugust 2022130 billion (dense)GLM-130B LicenseNo
BLOOM (language model)BigScience project (coordinated by Hugging Face)July 12, 2022176 billion (smaller variants 560M to 7.1B)OpenRAILWith conditions
No Language Left Behind (NLLB)Meta AIJuly 2022600M to 54.5BCC BY-NC 4.0No
OPT (Open Pre-trained Transformer)Meta AIMay 3, 2022125M to 175B (9 sizes)OPT-175B License AgreementNo
EXAONE (part of the family)LG AI Research (LG Corporation)December 14, 20211.2B to 32B denseApache 2.0Yes
Sentence-transformers/all-MiniLM-L6-v2 modelSentence Transformers project (2021 Hugging Face JAX/Flax community sprint)August 2021~22.7 millionApache 2.0Yes
Sentence-transformers/all-mpnet-base-v2 modelSentence Transformers project (2021 Hugging Face Flax/JAX community sprint)August 2021~110 millionApache 2.0Yes
SPLADENaver Labs EuropeJuly 12, 2021Not statedCC BY-NC-SA 4.0No
GPT-JEleutherAIJune 9, 20216,053,381,344 (~6.05B)Apache 2.0Yes
GLM (part of the family)Tsinghua University and Zhipu AI (Z.ai)June 2021110M to 744B total (40B active, GLM-5)MITYes
Huawei PanGu (part of the family)Huawei (Huawei Cloud and Noah's Ark Lab)April 20211.3B to 1.085T across the familyopenPangu Model License Agreement 1.0With conditions
Switch TransformerGoogle (Google Brain)January 11, 2021Up to 1.571 trillion (Switch-C, 2,048 experts)Apache 2.0Yes
mT5Google ResearchOctober 22, 2020300M to 13B (five sizes)Apache 2.0Yes
BlenderBotMeta AI (FAIR)April 29, 2020Up to 175B (BlenderBot 3)MITYes
ColBERTStanford University (Omar Khattab, Matei Zaharia)April 27, 2020110M (ColBERTv2)MITYes
LongformerAllen Institute for AIApril 10, 2020Not statedApache 2.0Yes
ELECTRAStanford University and Google BrainMarch 23, 202014M to 335MApache 2.0Yes
DeBERTaMicrosoft Research202022M backbone (v3-XSmall) to 1.5B (v2-XXLarge)MITYes
OPUS-MTLanguage Technology Research Group, University of Helsinki2020Not statedCC BY 4.0Yes
LayoutLMMicrosoft ResearchDecember 31, 2019~113M (BASE) to ~343M (LARGE), v1MITYes
Sentence-BERT (SBERT)Nils Reimers and Iryna Gurevych, UKP Lab, TU DarmstadtNovember 2019Not statedApache 2.0Yes
XLM-RoBERTaFacebook AI Research (Meta AI)November 2019~270M (Base); ~550M (Large); XL ~3.5B; XXL ~10.7BMITYes
T5 (language model)GoogleOctober 23, 201960 million to 11 billion (five sizes)Apache 2.0Yes
BART (language model)Facebook AI ResearchOctober 2019~140M (base); ~406M (large)Apache 2.0Yes
DistilBERTHugging FaceOctober 201966 millionApache 2.0Yes
ALBERTGoogle Research and Toyota Technological Institute at ChicagoSeptember 26, 201912M to 235MApache 2.0Yes
RoBERTaFacebook AI Research; University of WashingtonJuly 26, 2019125M (base), 355M (large)MITYes
XLNetCarnegie Mellon University and Google BrainJune 2019110M (Base), 340M (Large)Apache 2.0Yes
Baidu ERNIE (part of the family)Baidu, Inc.April 19, 2019Up to 424B total (47B active) in open-source ERNIE 4.5ERNIE 5.5 Flash Base Model License Agreement v1.0With conditions
GPT-2OpenAIFebruary 14, 20191.5B (XL); also 124M, 355M, 774M variantsModified MITYes
CamemBERTInria (Almanach), Facebook AI Research, Sorbonne Universite2019110M (base); 335M (large)MITYes
SciBERTAllen Institute for AI (AI2)2019~110MApache 2.0Yes
BERTGoogle AI LanguageOctober 11, 2018110M (Base); 340M (Large)Apache 2.0Yes
Bert-base-uncased modelGoogle AI LanguageOctober 2018~110 millionApache 2.0Yes
GPT (part of the family)OpenAIJune 11, 2018117M (GPT-1) to 175B (GPT-3); undisclosed from GPT-4 onwardModified MITYes
GPT-1OpenAIJune 11, 2018117 millionMITYes
ELMo (Embeddings from Language Models)Allen Institute for AI and University of WashingtonFebruary 2018~93.6 millionApache 2.0Yes

Expanded article table

Undated

ModelDeveloperReleasedParametersWeights licenseCommercial use
Ling-3.1-flashInclusionAI (Ant Group)Not statedAbout 560B total, 25B active (publisher's count)MITYes
Nemotron-Labs-TwoTowerNVIDIANot statedAbout 60B total (two 30B towers)NVIDIA Open Model LicenseWith conditions

Expanded article table

Image models

ModelDeveloperReleasedParametersWeights licenseCommercial use
Qwen-Image-2.1-TurboQwen team (Alibaba)October 9, 20267B (visual generation component)Qwen Research LicenseNo
Ming-Image-0.1-DesigninclusionAI (Ant Group)September 22, 20266B diffusion transformer per model (developer figure)MITYes
Qwen-Image-2.1Qwen team, AlibabaSeptember 20, 20267B diffusion transformer; 8B Qwen3-VL text encoderQwen Research LicenseNo
FLUX.2 (part of the family)Black Forest LabsNovember 25, 2025~32B (dev, pro, flex, max); klein 4B and 9BFLUX Non-Commercial LicenseNo
Hunyuan Image 3.0TencentSeptember 28, 2025~80B total; ~13B active per token; 64 expertsTencent Hunyuan Community LicenseWith conditions
HiDreamHiDream.aiApril 2025About 17 billionMITYes
Stable Diffusion 3.5Stability AIOctober 22, 20248.1B (Large); ~2.5B (Medium)Stability AI Community LicenseWith conditions
Flux (text-to-image model) (part of the family)Black Forest LabsAugust 1, 202412B (FLUX.1); 32B (FLUX.2); 4B and 9B (FLUX.2 klein)FLUX Non-Commercial LicenseNo
FLUX.1 (part of the family)Black Forest LabsAugust 1, 2024~12BFLUX Non-Commercial LicenseNo
Stable Diffusion 3 (part of the family)Stability AIJune 12, 2024800M to 8.1B (SD 3.5 Large)Stability AI Community LicenseWith conditions
Latent Consistency Models (LCM)Tsinghua University (IIIS)October 6, 2023Not statedOpenRAILWith conditions
SDXL (Stable Diffusion XL)Stability AIJuly 26, 2023~2.6B U-Net; ~6.6B total with refinerOpenRAILWith conditions
WürstchenPablo Pernias, Dominic Rampas, Mats L. Richter, Christopher J. Pal, Marc AubrevilleJune 1, 2023Stage A 18M; Stage B ~1B; Stage C ~0.99BMITYes
ControlNetLvmin Zhang, Anyi Rao and Maneesh Agrawala (Stanford University)February 10, 2023Not statedOpenRAILWith conditions
Stable DiffusionStability AI, with CompVis (LMU Munich), Runway, LAION, and EleutherAIAugust 22, 2022860M U-Net (SD 1.x); ~3.5B (SDXL base); 8.1B (SD 3.5 Large)Stability AI Community LicenseWith conditions
GLIDE (OpenAI) (part of the family)OpenAIDecember 20, 20213.5B (full model); ~300M released filtered modelMITYes
StyleGANNVIDIA ResearchDecember 12, 2018Not statedCC BY-NC 4.0No
CycleGANBerkeley AI Research (BAIR) Lab, UC BerkeleyMarch 2017Not statedBSD 3-ClauseYes

Expanded article table

Video models

ModelDeveloperReleasedParametersWeights licenseCommercial use
Pika H3 FlexPika (Mellis, Inc.)October 29, 202633B base generatorMiniMax H3 Community License AgreementWith conditions
SenseNova 6.8 Code2VideoSenseTime (SenseNova team)October 202627B (27,356,728,560)Apache 2.0Yes
MAGI-2 PreviewSand.aiAugust 5, 2026About 114B total, about 6B activeApache 2.0Yes
MiniMax H3MiniMaxJuly 31, 202633B (H3-Omni-Transformer generator)MiniMax H3 Community License AgreementWith conditions
Wan 2.1-VACEAlibaba (Tongyi Lab)May 14, 20251.3B and 14B (two variants)Apache 2.0Yes
Wan 2.1Alibaba (Tongyi Wanxiang team)February 25, 20251.3B and 14B variantsApache 2.0Yes
Step-VideoStepFun (Shanghai Jieyue Xingchen)February 17, 202530B (DiT backbone)MITYes
HunyuanVideoTencentDecember 3, 202413B (1.0); 8.3B (1.5)Tencent Hunyuan Community LicenseWith conditions
LTX-VideoLightricksNovember 21, 2024~1.9B (base); 13B (LTXV-13B)LTXV Open Weights LicenseWith conditions
Mochi 1Genmo Inc.October 22, 202410 billion (DiT) plus 362 million (VAE)Apache 2.0Yes
Hailuo AI (part of the family)MiniMaxSeptember 2024Not statedMiniMax H3 Community License AgreementWith conditions
CogVideoXZhipu AI and Tsinghua University (THUDM)August 6, 20242B or 5B (by variant)CogVideoX LicenseWith conditions
Open-SoraHPC-AI Tech (Colossal-AI team)March 17, 202411B (v2.0)Apache 2.0Yes
Pika (video generation) (part of the family)Pika LabsNovember 28, 2023Not statedMiniMax H3 Community License AgreementWith conditions
Stable Video DiffusionStability AINovember 21, 2023Not statedStability AI Community LicenseWith conditions

Expanded article table

Speech and audio models

ModelDeveloperReleasedParametersWeights licenseCommercial use
Nemotron 3.5 TranscribeNVIDIAOctober 28, 2026About 34B total, about 4.3B activeOpenMDW 1.1Yes
Whistle (Cactus Compute)Cactus ComputeOctober 2, 2026Not statedApache 2.0Yes
Cohere TranscribeCohere and Cohere LabsMarch 26, 2026Not statedApache 2.0Yes
Omnilingual ASRMeta FAIRNovember 10, 20250.32B to 7.81BApache 2.0Yes
VoxtralMistral AIJuly 15, 2025Voxtral Small 24.3B; Voxtral Mini 4.7BApache 2.0Yes
Sesame CSM (part of the family)Sesame AIMarch 13, 20251B backbone + 100M audio decoder (CSM-1B)Apache 2.0Yes
GLM-4-VoiceZhipu AI and Tsinghua University KEGOctober 25, 20249B (GLM-4-9B backbone)glm-4-voice LicenseWith conditions
SpiRit-LMMeta AI (FAIR)October 18, 20247B (Base and Expressive versions)FAIR Noncommercial Research LicenseNo
F5-TTSShanghai Jiao Tong University X-LANCE Lab, University of Cambridge, GeelyOctober 9, 2024335.8 millionCC BY-NC 4.0No
Qwen2-AudioQwen team, Alibaba CloudAugust 2024About 8.2BApache 2.0Yes
MoshiKyutaiJuly 3, 20247.7B total (Helium backbone is 7B)CC BY 4.0Yes
CosyVoiceTongyi SpeechTeam, Alibaba Group (FunAudioLLM project)July 2024300M (v1)Apache 2.0Yes
NVIDIA CanaryNVIDIAApril 2024182M to 2.5B across variantsCC BY 4.0Yes
NVIDIA ParakeetNVIDIA (first models with Suno.ai)December 2023110M to 1.1BCC BY 4.0Yes
XTTS (Coqui XTTS)Coqui AISeptember 30, 2023About 750MCoqui Public Model License 1.0.0No
SeamlessM4TMeta AIAugust 22, 20231.2B (Medium), 2.3B (Large and v2)CC BY-NC 4.0No
MusicGenMeta AI (FAIR)June 8, 2023300M (Small), 1.5B (Medium), 3.3B (Large)CC BY-NC 4.0No
Massively Multilingual Speech (MMS)Meta AIMay 22, 2023317M (MMS-300M), 965M (MMS-1B)CC BY-NC 4.0No
EnCodecMeta AI (FAIR)October 24, 2022Not statedMITYes
WhisperOpenAISeptember 21, 202239M (tiny) to 1,550M (large)MITYes
HuBERTMeta AI (then Facebook AI Research)June 14, 2021About 95M (Base), 317M (Large), 964M (X-Large)Apache 2.0Yes
Wav2Vec 2.0Meta (Facebook AI Research)June 20, 2020About 95M (BASE), about 317M (LARGE)Apache 2.0Yes
Jukebox (OpenAI)OpenAIApril 30, 2020About 1B (each upsampling prior) to about 5BOpenAI Jukebox Noncommercial Use LicenseNo
Wav2VecMeta AI (formerly Facebook AI Research)2019About 34M (wav2vec 1.0) to 2B (XLS-R 2B)Apache 2.0Yes
Ming-3.0inclusionAINot stated7.9B total, about 1.3B active (tts-tiny)MITYes

Expanded article table

Robotics and physical AI models

ModelDeveloperReleasedParametersWeights licenseCommercial use
AW-0Researchers at UC Berkeley, NVIDIA and the University of PennsylvaniaOctober 21, 2026Not statedBSD 2-ClauseYes
LingBot-VLA 2.0Robbyant (Ant Group)July 20266BApache 2.0Yes
NVIDIA Alpamayo 2 SuperNVIDIAJune 1, 202634 billionOpenMDW 1.1Yes
NVIDIA Cosmos 3NVIDIAJune 1, 20264B (Edge), 16B (Nano), 64B (Super)OpenMDW 1.1Yes
NVIDIA Isaac GR00T N1.7NVIDIAApril 17, 20263BNVIDIA Open Model LicenseWith conditions
Hy-EmbodiedTencent (Hunyuan team and Tencent Robotics X)April 2026About 4B (VLA-0.5), 6.2B (RxBrain-1.0)Apache 2.0Yes
NVIDIA Isaac GR00T N1.6NVIDIA (GEAR team)December 15, 20253BNVIDIA License (non-commercial)No
V-JEPA 2Meta AIJune 11, 20251.2B (world-model configuration; encoder over 1B)MITYes
SmolVLAHugging Face (LeRobot team)June 3, 2025~450M total (~100M action expert)Apache 2.0Yes
Isaac GR00T N1.5NVIDIAMay 2025~3B (GR00T-N1.5-3B)NVIDIA License (non-commercial)No
NVIDIA Cosmos ReasonNVIDIAMay 20257B (Cosmos-Reason1); 2B to 32B (Cosmos-Reason2)NVIDIA Open Model LicenseWith conditions
π0.5Physical IntelligenceApril 22, 2025~3.3B (~3B PaliGemma backbone + ~300M action expert)Gemma Terms of UseWith conditions
NVIDIA Isaac GR00T N1NVIDIA (GEAR research lab)March 18, 2025~2B (base N1)NVIDIA License (non-commercial)No
NVIDIA COMPASSNVIDIAFebruary 2025Not statedNVIDIA Open Model LicenseWith conditions
NVIDIA CosmosNVIDIAJanuary 6, 20254B to 14B (Cosmos 1.0 models)OpenMDW 1.1Yes
Meta MotivoMeta AI (FAIR)December 12, 2024~24.5M (S variants); ~288M (metamotivo-M-1)CC BY-NC 4.0No
π0Physical IntelligenceOctober 31, 2024~3.3B total (3B VLM backbone + ~300M action expert)Gemma Terms of UseWith conditions
OpenVLAStanford, UC Berkeley, TRI, Google DeepMind, Physical Intelligence, MITJune 20247BLlama Community LicenseWith conditions
Octo (Robot Policy)Octo Model TeamNot stated27M (Small) and 93M (Base)MITYes

Expanded article table

Science models

ModelDeveloperReleasedParametersWeights licenseCommercial use
Hydrogen-1.0Deep PrincipleNovember 2, 2026About 753B (753,329,940,480 in the BF16 checkpoint)MITYes
Ling-3.0-flash-SanteinclusionAISeptember 4, 2026124B total, 5.1B active per tokenMITYes
NVIDIA STRATANVIDIA, with U.S. national laboratoriesJune 30, 2026331.7MOpenMDW 1.1Yes
rBio (CZI)Chan Zuckerberg Initiative (CZI)August 21, 20253 billion (base model)Qwen Research LicenseNo
UMA (Universal Model for Atoms)FAIR Chemistry team, Meta AIJune 30, 2025UMA-S ~150M, UMA-M ~1.4B, UMA-L ~700M (total)FAIR Chemistry License v1With conditions
AlphaGenomeGoogle DeepMindJune 25, 2025Not statedAlphaGenome Model TermsNo
State (Arc Institute virtual cell model)Arc InstituteJune 23, 2025SE-600M checkpoint about 0.7BArc Research Institute State Model Non-Commercial LicenseNo
Boltz-2MIT (Jameel Clinic, CSAIL) and RecursionJune 2025Not statedMITYes
MedGemmaGoogleMay 20, 20254B and 27BHealth AI Developer Foundations Terms of UseWith conditions
TxGemmaGoogleMarch 25, 20252B, 9B, 27BHealth AI Developer Foundations Terms of UseWith conditions
Evo 2 (genomic model)Arc Institute, NVIDIA, Stanford, and collaboratorsFebruary 19, 2025About 40 billion (largest); also 7B and 1BApache 2.0Yes
MatterGenMicrosoft Research, AI for ScienceJanuary 16, 2025Not statedMITYes
GenCastGoogle DeepMindDecember 4, 2024Not statedCC BY 4.0Yes
BoltzMIT (CSAIL and Jameel Clinic)November 1, 2024Not statedMITYes
Chai-1Chai DiscoverySeptember 9, 2024Not statedApache 2.0Yes
ESM3 (part of the family)EvolutionaryScaleJune 25, 202498B flagship; also 1.4B and 7BMITYes
AlphaFold 3Google DeepMind and Isomorphic LabsMay 8, 2024Not statedAlphaFold 3 Model Parameters Terms of UseNo
NVIDIA Earth-2NVIDIAMarch 18, 2024Not statedNVIDIA Open Model LicenseWith conditions
AlphaGeometry (part of the family)Google DeepMindJanuary 17, 2024Up to 3.3B (AlphaGeometry 2 language model)CC BY 4.0Yes
GraphCastGoogle DeepMindNovember 14, 202336.7MCC BY 4.0Yes
RFdiffusionBaker Lab, UW Institute for Protein DesignMarch 30, 2023Not statedBSD 3-ClauseYes
Galactica (language model)Meta AI (Papers with Code team)November 15, 2022Five sizes, 125M to 120BCC BY-NC 4.0No
ESMFoldMeta AI (FAIR Protein Team)November 1, 2022Production checkpoints use 3B ESM-2 stemMITYes
BioGPTMicrosoft ResearchOctober 2022347M (base); 1.5B (BioGPT-Large)MITYes
OpenFoldAlQuraishi lab (Columbia University) and the OpenFold ConsortiumJune 2022Not statedCC BY 4.0Yes
AlphaFold-MultimerGoogle DeepMindOctober 4, 2021Not statedCC BY 4.0Yes
BioBERTDMIS Lab, Korea UniversityJanuary 25, 2019110M (Base); 340M (Large)Apache 2.0Yes

Expanded article table

Vision and multimodal models

ModelDeveloperReleasedParametersWeights licenseCommercial use
WeVisOne-VLTencent (WeChat Vision team)October 22, 20262B and 4B variants, each with a 9.6M pixel decoderApache 2.0Yes
Intern-DecisionInternLM team, Shanghai AI LaboratorySeptember 26, 20260.8B, 2B and 4BApache 2.0Yes
WeMM-EmbeddingTencent (WeChat Vision team)August 25, 2026Not statedApache 2.0Yes
HY-World 2.0Tencent HunyuanApril 16, 2026About 425M to 80B per componentTencent Hunyuan Community LicenseWith conditions
Lyra 2.0NVIDIA Spatial Intelligence LabApril 15, 2026Not statedNVIDIA Internal Scientific R&D Model LicenseNo
DeepSeek-OCRDeepSeekOctober 20, 2025About 3B decoderMITYes
DINOv3Meta AI (FAIR)August 14, 202521M to 6.7B (ViT-7B/16 largest)DINOv3 LicenseWith conditions
HunyuanWorld 1.0Tencent (Hunyuan team)July 26, 2025Not statedTencent Hunyuan Community LicenseWith conditions
Qwen2.5-VLAlibaba Cloud (Qwen team)January 26, 20253B, 7B, 32B, 72BApache 2.0Yes
DeepSeek-VL2DeepSeek (DeepSeek-AI)December 13, 20243B/16B/27B total; 1.0B/2.8B/4.5B activeDeepSeek Model LicenseWith conditions
DeepSeek JanusDeepSeekOctober 17, 20241.3B (Janus, JanusFlow); 1B and 7B (Janus-Pro)DeepSeek Model LicenseWith conditions
MolmoAllen Institute for AI (Ai2)September 25, 202472B, 7B (x2), MolmoE-1B (~1.5B active, 7.2B total)Apache 2.0Yes
Qwen2-VLAlibaba Cloud (Qwen team)August 29, 20242B, 7B, 72BApache 2.0Yes
Sapiens (computer vision)Meta (Reality Labs)August 22, 20240.3B, 0.6B, 1B, 2BCC BY-NC 4.0No
SAM 2Meta AI (FAIR)July 29, 2024Tiny 38.9M, Small 46M, Base+ 80.8M, Large 224.4MApache 2.0Yes
Chameleon (Meta AI)Meta AI (FAIR)June 18, 20247B and 34BChameleon Research LicenseNo
Florence-2Microsoft ResearchJune 16, 2024232M (base); 771M (large)MITYes
PaliGemmaGoogleMay 14, 20243B (PaliGemma); 3B, 10B and 28B (PaliGemma 2)Gemma Terms of UseWith conditions
DeepSeek-VLDeepSeekMarch 11, 20241.3B and 7BDeepSeek Model LicenseWith conditions
V-JEPAMeta AI (FAIR)February 15, 2024Up to ~630M (ViT-H/16 encoder)CC BY-NC 4.0No
MiniCPM-VOpenBMB (THUNLP and ModelBest Inc.)February 1, 20241.3B to 9B, depending on versionApache 2.0Yes
InternVLOpenGVLab (Shanghai AI Laboratory) and academic partnersDecember 21, 20230.94B to 240.7B (28B active in InternVL3.5-241B-A28B)Apache 2.0Yes
CogVLMZhipu AI and Tsinghua University (KEG)October 5, 2023About 17B (about 10B vision, 7B language)CogVLM LicenseWith conditions
Nougat (model)Meta AIAugust 25, 2023250M (small); 350M (base)CC BY-NC 4.0No
Qwen-VL (part of the family)Alibaba Cloud (Qwen team)August 22, 20239.6BQwen License (Tongyi Qianwen)With conditions
HieraMeta AI (FAIR)June 1, 202328M to 673MCC BY-NC 4.0No
ImageBindMeta AI (FAIR)May 9, 2023Not statedCC BY-NC-SA 4.0No
DINOv2Meta AI Research (FAIR)April 14, 202321M to 1.1B (ViT-S/14, ViT-B/14, ViT-L/14, ViT-g/14)Apache 2.0Yes
Segment Anything Model and Dataset (SAM and SA-1B)Meta AIApril 5, 2023~91M (ViT-B), ~308M (ViT-L), ~636M (ViT-H)Apache 2.0Yes
DINO (computer vision)Meta AIApril 202321M (ViT-S) to 6.7B (DINOv3 ViT-7B)DINOv3 LicenseWith conditions
LLaVA (Large Language and Vision Assistant)University of Wisconsin-Madison, Microsoft Research and Columbia UniversityApril 20237B and 13B backbones (LLaVA-1.5)Apache 2.0Yes
SigLIPGoogle DeepMindMarch 27, 202386M to 1B (SigLIP 2, ViT-B to ViT-g)Apache 2.0Yes
YOLOv8UltralyticsJanuary 10, 20233.2M (YOLOv8n) to 68.2M (YOLOv8x) for detectionGNU AGPL-3.0Yes
ConvNeXtFacebook AI Research (FAIR) and UC BerkeleyJanuary 20233.7M (V2-Atto) to 660M (V2-Huge)CC BY-NC 4.0No
I-JEPAMeta AIJanuary 2023About 632M (ViT-Huge)CC BY-NC 4.0No
MetaCLIPMeta AI2023Not statedCC BY-NC 4.0No
InternVideoOpenGVLab, Shanghai AI Laboratory; Nanjing University; SIAT, CASDecember 6, 2022Up to 6B (InternVideo2-6B)Apache 2.0Yes
Donut (Model)NAVER CLOVANovember 2021143M (176M when English token embeddings are counted)MITYes
CLIP (Contrastive Language-Image Pre-training)OpenAI2021Not statedMITYes
Swin TransformerMicrosoft Research Asia202129M (Swin-T) to 197M (Swin-L); about 3B (SwinV2-G)MITYes
DeiTFacebook AI Research (FAIR) and Sorbonne UniversityDecember 23, 20205M to 87MApache 2.0Yes
DETRFacebook AI Research (FAIR)202041M (ResNet-50) to 60M (ResNet-101)Apache 2.0Yes
EfficientNetGoogle Brain (Mingxing Tan, Quoc V. Le)20195.3M (B0) to 66M (B7)Apache 2.0Yes
ResNetKaiming He, Xiangyu Zhang, Shaoqing Ren and Jian Sun201625.6M (ResNet-50, TorchVision count)MITYes
YOLO (object detection)Joseph Redmon and co-authors (v1 to v3); later versions by Ultralytics and other groupsJune 2015Varies by version; 55.7M for YOLO26xGNU AGPL-3.0Yes
GoogLeNet (Inception v1)Google ResearchSeptember 17, 2014~5M (~6.8M counting all layers incl. auxiliary heads)Apache 2.0Yes
VGGNetVisual Geometry Group, University of OxfordSeptember 4, 2014138.4M (VGG-16); 143.7M (VGG-19)CC BY 4.0Yes
d3-turbovLLM Semantic Router teamNot stated26.1B loadedApache 2.0Yes

Expanded article table

Other models

ModelDeveloperReleasedParametersWeights licenseCommercial use
Decision 2.0vLLM Semantic Router projectOctober 3, 20260.60B to 29.37B (six checkpoints, nominal sizes 0.6B to 27B)Apache 2.0Yes
Extropic Z1TExtropicSeptember 4, 2026Not statedLicense not statedNot stated
Frontis-MA1FrontisAI (Horizon Research of Frontis.AI with Tsinghua University)July 30, 202635B total, about 3B active (Frontis-MA1-35B)CC BY-NC 4.0No
NVIDIA IsingNVIDIAApril 14, 202631B (Calibration 1.5); 35B total, 3B active (Calibration 1)OpenMDW 1.1Yes
Xiaomi MiMo-V2.5XiaomiApril 2026310B total, 15B active (standard)MITYes
Qwen-ImageQwen team, Alibaba (Tongyi Lab)August 4, 2025~20 billion (MMDiT backbone)Apache 2.0Yes
Tülu 3Allen Institute for AI (Ai2)November 21, 20248B, 70B, 405BLlama Community LicenseWith conditions
Hunyuan 3D (part of the family)TencentNovember 2024Up to ~10B shape model (2.5); 3.3B shape + 2B paint in 2.1Tencent Hunyuan Community LicenseWith conditions
Show-oShow Lab, National University of Singapore (with ByteDance co-authors)August 21, 2024About 1.3BApache 2.0Yes
ShieldGemmaGoogleJuly 31, 20242B, 9B, 27B (text); 4B (ShieldGemma 2)Gemma Terms of UseWith conditions
BGE (BAAI General Embedding)Beijing Academy of Artificial Intelligence (BAAI)August 2, 202324M to 326M (v1.5 embedders)MITYes
Shap-EOpenAIMay 3, 2023Not statedMITYes
Alpaca (model)Stanford CRFMMarch 13, 20237B (13B variant documented)CC BY-NC 4.0No
RWKVBo Peng (BlinkDL) and the RWKV project, hosted by the Linux FoundationAugust 2021169M to 14BApache 2.0Yes
AlphaFoldGoogle DeepMindJuly 15, 2021Not statedAlphaFold 3 Model Parameters Terms of UseNo
DistilGPT2Hugging FaceOctober 3, 201982M (81,912,576 learned parameters)Apache 2.0Yes

Expanded article table

Continue reading

More from See also

More to read

Topics:AI ModelsOpen Source AI

References

  1. ^Open Source Initiative. "The Open Source AI Definition 1.0". opensource.org/...open-source-ai-definition
  2. ^Apache Software Foundation. "Apache License, Version 2.0". January 2004. apache.org/...LICENSE-2.0
  3. ^Open Source Initiative. "The MIT License". opensource.org/...mit
  4. ^Creative Commons. "Attribution 4.0 International (CC BY 4.0)" (deed). creativecommons.org/...4.0
  5. ^Creative Commons. "Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)" (deed). creativecommons.org/...4.0

Improve this article

Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.

1 revision · v1 · 10,654 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Independent fact-check on November 17, 2026: the prose and its counts were checked against the cited sources and the table, and 91 rows were checked against official repositories; the license entries of 14 families and multi-size releases were corrected as a result. 54 further rows were added the same day from official repositories without a second independent check.

Cite this page: AI Wiki. "List of open-weight models." aiwiki.ai, updated 17 Nov 2026, fact-checked 17 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/list_of_open_weight_models

Edit article