Open-weight AI model origins by organization country, 2024-2025

Major open-weight model families by lead organization country, 2024-2025, family counts.

Hugging Face model/dataset metadata
Technology & AI
Open-weight AI model origins by organization country, 2024-2025: Major open-weight model families by lead organization country, 2024-2025, family counts.

What the Data Shows

Open-weight AI model origins are compared by lead organization country for 10 major model families released in 2024-2025. The chart uses five country rows, with counts of model families as the plotted value and shares measured against the 10-family scope. The United States has the largest count at 5 families, while China has 2 families. The United States and China together account for 7 of the 10 families, and the country assignment follows the lead releasing organization rather than every contributor affiliation.

Upgrade to Unbranded Graphics

Data table

Unit: families · Period: 2024-2025

RankEntryValuePeriod
1United States52024-2025
2China22025
3France12025
4United Arab Emirates12024
5Canada12024

Sources

Source 1: Meta

Publisher: Meta

Dataset/table: Llama 4 license/model card

Units: licence and modality disclosure

Date checked: 2026-05-24

Evidence: Supports Llama 4 as a Meta open-weight family under the Llama 4 Community License.

Source 2: Google DeepMind / Google

Publisher: Google DeepMind / Google

Dataset/table: Gemma 3 model card

Units: licence and modality disclosure

Date checked: 2026-05-24

Evidence: Supports Gemma 3 as a Google open-weight family governed by Gemma Terms of Use.

Source 3: Microsoft

Publisher: Microsoft

Dataset/table: Phi-4-multimodal-instruct model card

Units: licence and modality disclosure

Date checked: 2026-05-24

Evidence: Supports Phi-4 multimodal as a Microsoft open-weight model family under the MIT license.

Source 4: IBM

Publisher: IBM

Dataset/table: Granite-3.3-8B-Instruct model card

Units: licence and model-purpose disclosure

Date checked: 2026-05-24

Evidence: Supports IBM Granite 3.3 as an Apache 2.0 language-model release.

Source 5: Allen Institute for AI

Publisher: Allen Institute for AI

Dataset/table: OLMo 2 model card

Units: licence and model-purpose disclosure

Date checked: 2026-05-24

Evidence: Supports OLMo 2 as an Apache 2.0 open model family.

Source 6: Mistral AI

Publisher: Mistral AI

Dataset/table: Mistral Small 3.1 model card

Units: licence and modality disclosure

Date checked: 2026-05-24

Evidence: Supports Mistral Small 3.1 as an Apache 2.0 multimodal model family.

Source 7: Qwen / Alibaba Cloud

Publisher: Qwen / Alibaba Cloud

Dataset/table: Qwen3 model card

Units: licence and model-purpose disclosure

Date checked: 2026-05-24

Evidence: Supports Qwen3 as an Alibaba/Qwen open-weight family under Apache 2.0.

Source 8: DeepSeek AI

Publisher: DeepSeek AI

Dataset/table: DeepSeek-R1 model card

Units: licence and model-purpose disclosure

Date checked: 2026-05-24

Evidence: Supports DeepSeek-R1 as a MIT-licensed open-weight reasoning model family.

Source 9: Technology Innovation Institute

Publisher: Technology Innovation Institute

Dataset/table: Falcon3 model card

Units: licence, training-data and model-purpose disclosure

Date checked: 2026-05-24

Evidence: Supports Falcon 3 as a TII open-weight family under the TII Falcon-LLM License.

Source 10: Cohere Labs

Publisher: Cohere Labs

Dataset/table: Aya Expanse 8B model card

Units: licence and modality disclosure

Date checked: 2026-05-24

Evidence: Supports Aya Expanse 8B as a Cohere Labs open-weight research release under a noncommercial licence.

Data notes & caveats

The country field uses the lead releasing organization. Shares are included as scoped support metrics, while the primary chart plots raw family counts.

Methodology

Model families were counted once by lead organization country using official model-card evidence. This avoids treating every variant or parameter size as a separate release.

Source consistency notes

Rows use the same 10-family scope and official model-card source family. The values are source-labelled because the scope is selected rather than a full institutional dataset.

Licensing

Branded MarketStats graphics are free to use*

Keep the visible MarketStats attribution, source line and canonical page link intact. For clean unbranded assets, commercial use or higher-volume reuse, choose one of the licensing tiers below.

CreatorUS$14.99/mo · 30/month · PNG/WebPRequest CreatorProUS$79.99/mo · 300/month · SVG includedRequest ProEnterpriseUS$379.99/mo · 1500/month · team accessRequest Enterprise
View tiers and terms
Request unbranded access

This specific graphic needs a source-rights review before an unbranded download. Branded use remains available.

* Free branded use is subject to the Terms and requires attribution/canonical-link preservation. MarketStats licenses the graphic presentation and MarketStats-created editorial layer, not the underlying third-party source data. Source attribution and source-provider obligations still apply.