Releases·Model Release
Saudi Arabia's HUMAIN unveils humain-m3, a 428B Arabic frontier model built by MiniMax
HUMAIN, a Saudi PIF (sovereign wealth fund) company, released humain-m3, a frontier Arabic model it commissioned and that MiniMax developed. The model is a 428B-parameter mixture-of-experts on the MiniMax-M3 lineage, with roughly a trillion additional Arabic pretraining tokens. It illustrates a sovereign-AI playbook: buying and adapting an existing frontier lineage rather than training a base model from scratch for one language.
777 engagement2 sourcesFri, Sep 4, 2026, 11:18 PM
AI Overview
- HUMAIN, a Saudi PIF company, released humain-m3 in research preview, a 428B mixture-of-experts model on the MiniMax-M3 lineage.
- The model was commissioned from MiniMax and trained with roughly a trillion extra tokens of Arabic pretraining.
- Weights ship next month under the MiniMax Community License, so licensing terms are set by MiniMax in Shanghai.
- HUMAIN's PR Newswire release includes benchmark numbers for the model.
Related
Developments·Benchmark
MiniMax-M3 tops agent benchmark at $0.018 per task
MiniMax-M3, a frontier lab model, completed a real-world agent benchmark task—spinning up an inbox and sending a business email end-to-end—for $0.018, the lowest cost on the board. The result highlights the model's efficiency for practical agentic workflows.
MiniMax-M3, a frontier lab model, completed a real-world agent benchmark task—spinning up an inbox and sending a business email end-to-end—for $0.018, the lowest cost on the board. The result highlights the model's efficiency for practical agentic workflows.
73 engagement1 sourcexTue, Aug 25, 2026, 11:51 PM
Releases·Model Release
Tencent Hy Team releases Hy3, a 295B-parameter MoE model
Tencent's Hy Team released Hy3, a 295-billion-parameter Mixture-of-Experts (MoE) language model with 21B active parameters and a 3.8B MTP layer. The model outperforms similar-sized models and rivals open-source models with 2-5x parameters. It is available on Hugging Face in full (598GB) and FP8 quantized (300GB) versions.
Tencent's Hy Team released Hy3, a 295-billion-parameter Mixture-of-Experts (MoE) language model with 21B active parameters and a 3.8B MTP layer. The model outperforms similar-sized models and rivals open-source models with 2-5x parameters. It is available on Hugging Face in full (598GB) and FP8 quantized (300GB) versions.
0 engagement1 sourcerssMon, Jul 6, 2026, 11:57 PM
Releases·Model Release
Motif Technologies releases Motif-3, a 314B MoE model with 256K context
Motif Technologies has released Motif-3, a 314-billion-parameter Mixture-of-Experts language model with a 256K context window. The model is built from the ground up, suggesting a new architecture rather than a fine-tune. This release marks a significant entry in the large MoE model space.
Motif Technologies has released Motif-3, a 314-billion-parameter Mixture-of-Experts language model with a 256K context window. The model is built from the ground up, suggesting a new architecture rather than a fine-tune. This release marks a significant entry in the large MoE model space.
0 engagement1 sourcerssWed, Aug 12, 2026, 04:51 AM
Releases·Model Release
Shanghai AI Lab releases Intern-S2-Preview-397B on HuggingFace
The Shanghai Artificial Intelligence Laboratory released Intern-S2-Preview-397B, a 397-billion-parameter model, on HuggingFace on July 18, 2026. The model is available for download and inference, continuing the Intern series of large language models.
The Shanghai Artificial Intelligence Laboratory released Intern-S2-Preview-397B, a 397-billion-parameter model, on HuggingFace on July 18, 2026. The model is available for download and inference, continuing the Intern series of large language models.
69 engagement1 sourceredditSat, Jul 18, 2026, 01:35 AM
Releases·Model Release
DFM releases Mimir v1, a 1B HRM model trained on permissible data
DFM released Mimir v1, a 1-billion-parameter language model based on the Hierarchical Reasoning Model (HRM) architecture, trained from scratch on 161 datasets using only permissible post-training data. It delivers competitive English performance and sets a new state of the art for Danish, offering an open alternative to models trained on non-permissible data.
DFM released Mimir v1, a 1-billion-parameter language model based on the Hierarchical Reasoning Model (HRM) architecture, trained from scratch on 161 datasets using only permissible post-training data. It delivers competitive English performance and sets a new state of the art for Danish, offering an open alternative to models trained on non-permissible data.
28 engagement2 sourcesarxiv, redditThu, Aug 13, 2026, 05:37 PM