United States
Open models from United States
We track 122 open-weight models from 7 labs in United States. They account for 46 of 205 tracked releases (22%) from Dec 2025 to Nov 2026. The newest is granite-switch-4.1-8b, listed on 14 Oct 2026.
Explore the models
Latest models
Newest listed firstCheapest tracked APIs
- gemma-4-26B-A4B-itPrimeIntellect · BF16$0.0208 (₹1.96)
- gpt-oss-20bPrimeIntellect · FP4$0.0225 (₹2.12)
- gemma-3-4b-itParasail · BF16$0.0425 (₹4.01)
- gpt-oss-120bPrimeIntellect · FP4$0.045 (₹4.24)
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16Darkbloom · INT4 (other providers mostly serve BF16)$0.0488 (₹4.60)
- gemma-3-12b-itDeepInfra · BF16 (other providers mostly serve INT4)$0.06 (₹5.66)
- phi-4DeepInfra · BF16$0.07 (₹6.60)
- Llama-3.2-1B-InstructCloudflare · precision not listed$0.0705 (₹6.65)
Per million tokens, blended three parts input to one part output. INR at ₹94.29 per dollar.
How the work developed
Meta released Llama base and chat models with downloadable weights under a custom community licence. The model card documents pretraining and dialogue fine-tuning separately. That approach lets developers adapt the weights while retaining model-specific usage conditions.
The Allen Institute for AI took a more reproducible research approach with OLMo, publishing training code and checkpoints alongside the Dolma training dataset. Google also released Gemma weights under its own terms, with model cards describing architecture, training and intended uses. These US lab entries cover several approaches to openness: access to weights, access to training materials, and differing permissions for reuse.
Sources and counting method
Counts cover selected repositories, including model variants. Dates come from Hugging Face repository creation dates and may differ from public announcements. Location follows the lab record, not the nationality of individual contributors.