Releases·Model Release
Tencent open-sources 780B-parameter MoE model with 1M-token context window
Tencent released an open-source 780B-parameter mixture-of-experts model. The model uses 256 experts, supports a 1-million-token context window, and weighs 814 GB even after FP8 quantization. The release gives practitioners access to frontier-scale open weights, though the size makes deployment demanding.
0 engagement1 sourceMon, Aug 31, 2026, 09:21 AM
AI Overview
- Model uses a mixture-of-experts architecture with 256 experts
- Supports a 1-million-token context window
- Weights total 814 GB even after FP8 quantization
- Coverage appears on Medium, emphasizing the architecture as the notable part
Entities
Related
Social·Community
Qwen3.8 'Opened' a 2.4-Trillion-Parameter
A model card on Hugging Face for Qwen/Qwen3.8–2.4T-A95B lists a 2.4-trillion-parameter count, which appears to be a typo. The card was fetched on August 16, 2026, and the discrepancy has drawn attention.
A model card on Hugging Face for Qwen/Qwen3.8–2.4T-A95B lists a 2.4-trillion-parameter count, which appears to be a typo. The card was fetched on August 16, 2026, and the discrepancy has drawn attention.
0 engagement1 sourcerssThu, Aug 20, 2026, 02:01 PM
Releases·Tool Release
Tencent's AngelSlim shrinks AI models via quantization, distillation, and speculative decoding
An Open Intelligence article examines Tencent's AngelSlim, an open toolkit for shrinking large AI models without major accuracy loss. It covers quantization, speculative decoding, sparse attention, token pruning, caching, and distillation as complementary compression techniques. For practitioners, it maps which methods fit inference-cost reduction across serving scenarios.
An Open Intelligence article examines Tencent's AngelSlim, an open toolkit for shrinking large AI models without major accuracy loss. It covers quantization, speculative decoding, sparse attention, token pruning, caching, and distillation as complementary compression techniques. For practitioners, it maps which methods fit inference-cost reduction across serving scenarios.
0 engagement1 sourcerssFri, Oct 2, 2026, 06:43 PM
Social·Community
Moonshot AI Releases Kimi K3, 2.8-Trillion-Parameter Open-Weight Model Claimed to Run on a 4GB GPU
Towards AI article claims Kimi K3, a 2.8T-parameter open-weight model, runs on a single 4GB GPU
Towards AI article claims Kimi K3, a 2.8T-parameter open-weight model, runs on a single 4GB GPU
0 engagement1 sourcerssFri, Sep 11, 2026, 10:31 PM
Releases·Model Release
ByteDance begins early training of 10-trillion-parameter model
ByteDance is at an early stage of training a model with as many as 10 trillion parameters, according to a report. This would make it one of the largest AI models ever attempted, potentially surpassing current state-of-the-art models in scale.
ByteDance is at an early stage of training a model with as many as 10 trillion parameters, according to a report. This would make it one of the largest AI models ever attempted, potentially surpassing current state-of-the-art models in scale.
681 engagement2 sourcesreddit, rssFri, Aug 7, 2026, 06:56 AM
Releases·Model Release
Moonshot AI releases Kimi K3, a 2.8-trillion-parameter open-weight model
Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-weight model, on Hugging Face at midnight UTC on July 27, 2026. The model reportedly narrows the frontier gap to 4 points, marking a significant advance in open-weight AI.
Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-weight model, on Hugging Face at midnight UTC on July 27, 2026. The model reportedly narrows the frontier gap to 4 points, marking a significant advance in open-weight AI.
0 engagement1 sourcerssMon, Jul 27, 2026, 12:00 AM