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.

889 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.

889

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

129

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.

889

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

256

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 (311) · MIT License (125) · Other (unclassified licence) (83) · Creative Commons Attribution-NonCommercial 4.0 (24) · Apple Sample Code / ML Research License (15) · Llama 3.2 Community License (12)

Category · this pageCommercial useRedistributionDerivativesHosting
Open source n=0————
Open weights n=508042804280428042
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

889 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
llama-3-1-nemotron-ultra-253b-v1-reasoningOpen weightsNVIDIA—unclassified—128K—gpqa-diamond#205ifbench#215humanitys-last-exam#2390——
llama-3-3-nemotron-super-49bOpen weightsNVIDIA—unclassified—128K—ifbench#206tau2-bench#209gpqa-diamond#2580——
hermes-4-llama-3-1-70bOpen weightsNous Research—unclassified—128K—terminal-bench#210humanitys-last-exam#221tau2-bench#2400——
llama-3-1-nemotron-instruct-70bOpen weightsNVIDIA—unclassified—128K—terminal-bench#210tau2-bench#232ifbench#2780——
magistral-smallOpen weightsMistral AI—unclassified—40K—terminal-bench#210tau2-bench#210humanitys-last-exam#2330——
olmo-2-32bOpen weightsAllen Institute for AI—unclassified—4.1K—ifbench#216terminal-bench#279tau2-bench#3020——
qwen3-1.7b-instructOpen weightsAlibaba Group—unclassified—32K—tau2-bench#217terminal-bench#279ifbench#2930——
Phi 4Open weightsMicrosoft · textMIT14.7B16.4K11 Dec 2024terminal-bench#222gpqa-diamond#295tau2-bench#3021from $0.14 out✓ fits8.9 GB4bit✓ fits17.4 GB8bit
g9v3-3bOpen weightsAI9Stars—unclassified—131.1K—artificial-analysis-intelligence-index#223humanitys-last-exam#348gpqa-diamond#3520——
llama-3-instruct-70bOpen weightsMeta AI—unclassified—8.19K—ifbench#223terminal-bench#265tau2-bench#3020——
NVIDIA-Nemotron-Nano-9B-v2Open weightsNVIDIA—unclassified—131.1K—tau2-bench#229terminal-bench#256ifbench#2900——
Command AOpen weightsCohere · text—unclassified—256K13 Mar 2025ifbench#230terminal-bench#265tau2-bench#2702from $10 out——
Llama-3.1-70BOpen weightsMeta AI—unclassified—128K—terminal-bench#233ifbench#245tau2-bench#2700——
granite-4-0-nano-1bOpen weightsIBM—unclassified—128K—tau2-bench#234terminal-bench#279ifbench#3230——
R1 Distill Llama 70BOpen weightsDeepSeek · text—unclassified—8.19K23 Jan 2025tau2-bench#239terminal-bench#256artificial-analysis-intelligence-index#2912from $0.80 out——
granite-4-0-h-smallOpen weightsIBM—unclassified—128K—terminal-bench#240tau2-bench#262ifbench#2700——
granite-4.1-3bOpen weightsIBM · textApache-2.03.4B131.1K6 Apr 2026terminal-bench#240ifbench#252tau2-bench#2530✓ fits2.5 GB4bit✓ fits4.4 GB8bit
jamba-1-7-largeOpen weightsAI21 Labs—unclassified—256K—terminal-bench#240ifbench#240tau2-bench#2770——
sarvam-m-reasoningOpen weightsSarvam—unclassified—32.8K—terminal-bench#240ifbench#266tau2-bench#3020——
Qwen3-0.6BOpen weightsQwen · textApache-2.0751.6M32K27 Apr 2025tau2-bench#245terminal-bench#279humanitys-last-exam#2800✓ fits0.9 GB4bit✓ fits1.4 GB8bit
exaone-4-0-1-2bOpen weightsLG AI Research · textOther1.28B64K11 Jul 2025tau2-bench#249humanitys-last-exam#273terminal-bench#2790✓ fits1.2 GB4bit✓ fits2.0 GB8bit
solar-miniOpen weightsUpstage—unclassified—4.1K—tau2-bench#250artificial-analysis-intelligence-index#3690——
Mistral Small 3Open weightsMistral AI · text—unclassified—32.8K30 Jan 2025tau2-bench#253ifbench#297gpqa-diamond#3482from $0.08 out——
granite-4-0-h-nano-1bOpen weightsIBM—unclassified—128K—tau2-bench#253terminal-bench#279ifbench#2990——
DeepSeek-R1-0528-Qwen3-8BOpen weightsDeepSeek · textMIT8.19B32.8K29 May 2025terminal-bench#256gpqa-diamond#273humanitys-last-exam#2750✓ fits5.2 GB4bit✓ fits9.9 GB8bit
Granite 4.0 MicroOpen weightsIBM · text—unclassified—131K20 Oct 2025terminal-bench#256tau2-bench#282ifbench#3051from $0.112 out——
granite-4-0-h-350mOpen weightsIBM—unclassified—32.8K—humanitys-last-exam#256tau2-bench#272terminal-bench#2790——
Qwen3-4BOpen weightsQwen · textApache-2.04.02B32K27 Apr 2025tau2-bench#257ifbench#261gpqa-diamond#3150✓ fits2.8 GB4bit✓ fits5.1 GB8bit
Olmo-3-7B-InstructOpen weightsAllen Institute for AI · textApache-2.07.3B65.5K19 Nov 2025ifbench#258humanitys-last-exam#278terminal-bench#2790✓ fits4.7 GB4bit✓ fits8.9 GB8bit
qwen3-4b-instructOpen weightsAlibaba Group—unclassified—32K—ifbench#261artificial-analysis-intelligence-index#326humanitys-last-exam#3560——
gemma-3n-e2bOpen weightsGoogle—unclassified—32K—terminal-bench#265tau2-bench#302ifbench#3190——
lfm2-2-6bOpen weightsLiquid AI—unclassified—32.8K—terminal-bench#265tau2-bench#277humanitys-last-exam#2870——
ling-mini-2-0Open weightsinclusionAI · textMIT16.3B131K8 Sept 2025terminal-bench#265tau2-bench#279gpqa-diamond#3000✓ fits9.8 GB4bit✓ fits19.2 GB8bit
DeepSeek-R1-Distill-Qwen-32BOpen weightsDeepSeek · textMIT32.8B128K20 Jan 2025gpqa-diamond#271artificial-analysis-intelligence-index#280ifbench#3140✓ fits19.3 GB4bit✓ fits38.2 GB8bit
jamba-1-7-miniOpen weightsAI21 Labs—unclassified—258K—ifbench#272terminal-bench#279tau2-bench#2820——
qwen3-0.6b-instructOpen weightsAlibaba Group—unclassified—32K—tau2-bench#272terminal-bench#279humanitys-last-exam#2800——
Reka Flash 3Open weightsrekaai · text—unclassified—65.5K12 Mar 2025terminal-bench#279ifbench#279tau2-bench#3021from $0.20 out——
apertus-70b-instructOpen weightsSwiss AI Initiative—unclassified—65.5K—terminal-bench#279tau2-bench#281humanitys-last-exam#2870——
apertus-8b-instructOpen weightsSwiss AI Initiative—unclassified—65.5K—terminal-bench#279tau2-bench#287humanitys-last-exam#3150——
gemma-3-1bOpen weightsGoogle—unclassified—32K—terminal-bench#279tau2-bench#293humanitys-last-exam#2950——
gemma-3-270mOpen weightsGoogle—unclassified—32K—terminal-bench#279tau2-bench#297ifbench#3340——
granite-3-3-8b-instructOpen weightsIBM—unclassified—128K—terminal-bench#279tau2-bench#293ifbench#3160——
granite-4-0-350mOpen weightsIBM—unclassified—32.8K—terminal-bench#279tau2-bench#279humanitys-last-exam#2850——
lfm2-8b-a1bOpen weightsLiquid AI—unclassified—32.8K—terminal-bench#279tau2-bench#293ifbench#2980——
llama-3-instruct-8bOpen weightsMeta AI—unclassified—8.19K—terminal-bench#279tau2-bench#302ifbench#3070——
olmo-2-7bOpen weightsAllen Institute for AI—unclassified—4.1K—terminal-bench#279humanitys-last-exam#291tau2-bench#3020——
phi-3-miniOpen weightsMicrosoft—unclassified—4.1K—terminal-bench#279tau2-bench#302ifbench#3090——
phi-4-miniOpen weightsMicrosoft—unclassified—128K—terminal-bench#279tau2-bench#299ifbench#3220——
llama-3-1-nemotron-nano-4b-reasoningOpen weightsNVIDIA—unclassified—128K—tau2-bench#286humanitys-last-exam#302ifbench#3020——
deepseek-coder-v2-liteOpen weightsDeepSeek—unclassified—128K—humanitys-last-exam#293gpqa-diamond#413artificial-analysis-intelligence-index#4270——

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