Skip to content
AI Atlas
Open model frontier

Open models

Every canonical model whose weights can be downloaded, described by measurable properties — licence permissions, parameters, context, release, benchmark ranks, providers, estimated local fit. Openness is a set of observed dimensions here, never a score.

870 downloadable models

Open sourceWeights and code published under an OSI-approved licence that allows commercial use and derivatives.

1

Open weightsWeights downloadable under a permissive or Creative Commons licence allowing commercial use; code or data may be missing.

870

Restricted weightsWeights downloadable, but the licence restricts commercial use, hosting, derivatives or field of use (community, research and RAIL licences).

79

Downloadable totalUniverse = canonical models whose weights can be downloaded (open-weights, open-source, restricted-weights). hardware_fit values are ESTIMATES (see /methodology); best_results are ranks inside each benchmark's primary comparability group — no composite score.

870

New in 30 daysDownloadable models with a release date inside the last 30 days (release_date claim; models without one are not counted).

269

Openness explorer

Categories and licence permissions

Left: the downloadable universe by openness category. Right: what the licences on this page allow, counted per category — allowed · restricted · unknown.

Top licences: Apache License 2.0 (275) · Other (unclassified licence) (115) · MIT License (92) · Creative Commons Attribution-NonCommercial 4.0 (22) · Apple Sample Code / ML Research License (15) · Llama 3.2 Community License (12)

Category · this pageCommercial useRedistributionDerivativesHosting
Open source n=0————
Open weights n=507043704370437043
restricted-weights n=0————

Counts read: allowed · restricted · unknown / unclassified — over the 50 models on this page, from each licence's stated terms (ontology).

Downloadable models

870 models · sorted by best benchmark rank

Permissions, in order:commercial use· redistribution· derivatives· hosting allowed restricted unknown

ModelLicence · permissionsParamsContextReleaseBest resultsProvidersFit 64 GB @4bit · 128 GB @8bit Estimated
diffusiongemma-26b-a4bOpen weightsGoogle—unclassified—256K—mmmu-pro#84ifbench#107humanitys-last-exam#1900——
hypernova-60bOpen weightsMultiverse Computing—unclassified—131.1K—ifbench#84terminal-bench#110tau2-bench#1270——
gpt-oss-20bOpen weightsOpenAI · textApache-2.020.9B131.1K4 Aug 2025scicode#86ifbench#88tau2-bench#1313from $0.13 out✓ fits12.5 GB4bit✓ fits24.6 GB8bit
Mistral Small 4Open weightsMistral AI · image, textApache-2.0—262.1K16 Mar 2026scicode#87mmmu-pro#115terminal-bench#1312from $0.30 out——
olmo-3-1-32b-instructOpen weightsAllen Institute for AI—unclassified—65.5K—ifbench#87tau2-bench#242humanitys-last-exam#2640——
k-exaoneOpen weightsLG AI Research—unclassified—256K—ifbench#89tau2-bench#112terminal-bench#1130——
deepseek-r1-0120Open weightsDeepSeek—unclassified—128K—scicode#90terminal-bench#198ifbench#2100——
granite-4.2-30bOpen weightsIBM—unclassified—131.1K—scicode#92artificial-analysis-intelligence-index#160humanitys-last-exam#1790——
minicpm5-1bOpen weightsOpenBMB—unclassified—128K—tau2-bench#92ifbench#150humanitys-last-exam#2550——
Qwen3 VL 32B InstructOpen weightsQwen · image, text—unclassified—131.1K23 Oct 2025mmmu-pro#94ifbench#108tau2-bench#1531from $0.416 out——
Qwen3.5-2BOpen weightsQwen—unclassified—262.1K—tau2-bench#94mmmu-pro#140terminal-bench#2220——
g9v3-39a5bOpen weightsAI9Stars—unclassified—131.1K—scicode#94artificial-analysis-intelligence-index#108humanitys-last-exam#1370——
nemotron-3-nano-omni-30b-a3bOpen weightsNVIDIA—unclassified—256K—ifbench#94mmmu-pro#122tau2-bench#1550——
Mistral Large 3Open weightsMistral AIApache-2.0—256K—scicode#95mmmu-pro#118terminal-bench#1411from $1.5 out——
nanbeige4-1-3bOpen weightsNanbeige—unclassified—256K—gpqa-diamond#96humanitys-last-exam#187ifbench#2380——
step-3-vl-10bOpen weightsStepFun—unclassified—65.5K—mmmu-pro#96ifbench#148humanitys-last-exam#1900——
tri-21b-think-v0-5Open weightsTrillion Labs—unclassified—32K—tau2-bench#96ifbench#126artificial-analysis-intelligence-index#2610——
Qwen3 Coder NextOpen weightsQwen · text—unclassified—262.1K4 Feb 2026scicode#98tau2-bench#101terminal-bench#1271from $0.80 out——
k2-think-v2Open weightsMBZUAI Institute of Foundation Models—unclassified—262.1K—ifbench#98terminal-bench#186humanitys-last-exam#2040——
DeepSeek V3.1Open weightsDeepSeek · text—unclassified840B163.8K5 Sept 2025terminal-bench#100humanitys-last-exam#153gpqa-diamond#1582from $0.95 out✗ too large483.5 GB4bit✗ too large966.5 GB8bit
apriel-v1-5-15b-thinkerOpen weightsServiceNow—unclassified—128K—ifbench#100mmmu-pro#114tau2-bench#1190——
deepseek-v3-2-0925Open weightsDeepSeek—unclassified—128K—terminal-bench#100artificial-analysis-intelligence-index#171ifbench#1790——
qwen3-32b-instructOpen weightsAlibaba Group—unclassified—32.8K—scicode#100ifbench#231terminal-bench#2330——
Qwen3 VL 30B A3B InstructOpen weightsQwen · image, text—unclassified—262.1K6 Oct 2025mmmu-pro#101ifbench#167terminal-bench#1981from $0.60 out——
longcat-flash-liteOpen weightsLongCat—unclassified—256K—tau2-bench#101terminal-bench#164ifbench#1800——
Qwen3 Next 80B A3B InstructOpen weightsQwen · text—unclassified—262.1K11 Sept 2025ifbench#103tau2-bench#161humanitys-last-exam#1641from $1.1 out——
qwen3-30b-a3b-2507Open weightsAlibaba Group—unclassified—262.1K—scicode#104ifbench#145humanitys-last-exam#1980——
Devstral Small 2Open weightsMistral AI—unclassified—256K—scicode#106terminal-bench#137mmmu-pro#1370——
k2-v2Open weightsMBZUAI Institute of Foundation Models—unclassified—512K—ifbench#106terminal-bench#170humanitys-last-exam#1950——
NVIDIA Nemotron 3.5 Lightning 30B A3BOpen weightsNVIDIA · text—unclassified—262.1K11 Aug 2026scicode#108artificial-analysis-intelligence-index#176gpqa-diamond#1923from $0.20 out——
qwen3-omni-30b-a3b-instructOpen weightsAlibaba Group—unclassified—65.5K—mmmu-pro#109ifbench#176gpqa-diamond#2080——
Granite 4.2 8BOpen weightsIBM · text—unclassified—131.1K31 Aug 2026scicode#110artificial-analysis-intelligence-index#205humanitys-last-exam#2091from $0.25 out——
Kimi K2 0905Open weightsMoonshot AI · textOther1.03T262.1K3 Sept 2025terminal-bench#110tau2-bench#113artificial-analysis-intelligence-index#1552from $2.5 out✗ too large590.7 GB4bit✗ too large1180.9 GB8bit
nvidia-nemotron-3-nano-4bOpen weightsNVIDIA—unclassified—262K—ifbench#110terminal-bench#186tau2-bench#1990——
Qwen3 14BOpen weightsQwen · text—unclassified—131.1K28 Apr 2025scicode#111tau2-bench#174ifbench#1981from $0.91 out——
qwen3-14b-instructOpen weightsAlibaba Group—unclassified—32.8K—scicode#111tau2-bench#184ifbench#1980——
solar-open-100b-reasoningOpen weightsUpstage—unclassified—128K—ifbench#113tau2-bench#147humanitys-last-exam#1960——
Mistral Small 3.2Open weightsMistral AI—unclassified—128K—scicode#114mmmu-pro#134terminal-bench#1860——
Qwen3 8BOpen weightsQwen · textApache-2.08.19B131.1K27 Apr 2025scicode#115tau2-bench#202terminal-bench#2401from $0.455 out✓ fits5.2 GB4bit✓ fits9.9 GB8bit
qwen3-8b-instructOpen weightsAlibaba Group—unclassified—32.8K—scicode#115tau2-bench#223terminal-bench#2400——
Qwen3 VL 8B InstructOpen weightsQwen · image, textApache-2.08.77B262.1K11 Oct 2025mmmu-pro#116tau2-bench#196ifbench#2021from $0.455 out✓ fits5.5 GB4bit✓ fits10.6 GB8bit
Mistral Small 3.1Open weightsMistral AI—unclassified—128K—scicode#117terminal-bench#182tau2-bench#2210——
minicpm5-2bOpen weightsOpenBMB—unclassified—131.1K—scicode#118artificial-analysis-intelligence-index#182humanitys-last-exam#2180——
Magistral Small 1.2Open weightsMistral AI—unclassified—128K—mmmu-pro#119ifbench#171tau2-bench#2020——
granite-4.2-3bOpen weightsIBM—unclassified—131.1K—scicode#120humanitys-last-exam#252artificial-analysis-intelligence-index#2560——
GLM 4.5 AirOpen weightsZ.ai (Zhipu AI) · textMIT110.5B131.1K20 Jul 2025terminal-bench#121tau2-bench#151gpqa-diamond#2002from $0.85 out✗ too large64.0 GB4bit✗ too large127.5 GB8bit
ling-3-0-tinyOpen weightsinclusionAI—unclassified—262.1K—scicode#121gpqa-diamond#199artificial-analysis-intelligence-index#2010——
Ministral 3 14BOpen weightsMistral AIApache-2.0—256K—scicode#122mmmu-pro#130tau2-bench#2071from $0.20 out——
LFM2.5-8B-A1BOpen weightsLiquid AIOther8.47B1B active32.8K28 May 2026ifbench#123terminal-bench#210humanitys-last-exam#2490✓ fits5.4 GB4bit✓ fits10.2 GB8bit
Llama 4 ScoutOpen weightsMeta AI · image, textOther108.6B1.31M2 Apr 2025mmmu-pro#124scicode#125ifbench#2051from $0.30 out✗ too large63.0 GB4bit✓ fits125.4 GB8bit

Best results = rank inside each benchmark's primary comparability group; several dimensions are shown side by side and never combined. Fit columns are estimates (64 GB device at 4-bit, 128 GB at 8-bit, 8K context) — method on /methodology. unclassified = the raw licence label is not yet mapped in the ontology.

MethodUniverse = canonical models whose weights can be downloaded (open-weights, open-source, restricted-weights). hardware_fit values are ESTIMATES (see /methodology); best_results are ranks inside each benchmark's primary comparability group — no composite score. /methodology