Termsslug: sparse-activation

Sparse Activation

稀疏激活

A mechanism that activates only a subset of parameters or sub-networks during inference to reduce computation

Curated Intelligence

12 items · Page 1/2
IT之家 · 智能时代T2·Open Source
78 pts
Tencent Open-Sources Hy4 flash with 224B Total Parameters

Tencent Hunyuan has officially open-sourced the Hy4 flash family, including the standard, FP8, and Base variants, available on GitHub, Hugging Face, ModelScope, and GitCode. The model carries 224B total parameters (including 64B Engram parameters) while activating only 12B per token, supports multimodal input and a 1M-token context window, and is positioned as the most cost-effective Hy model to date, already deployed across Tencent Cloud TokenHub, Yuanbao, WeCom, and other products. Tencent also teased that the full Hy4 model is being open-sourced soon.

Arena.ai (@arena)T1·Models
78 pts
Baidu Releases ERNIE 5.5 Flash Preview with 280B Open Weights

Baidu unveiled ERNIE 5.5 Flash Preview, built on a new architecture optimized for efficient inference with 280B total parameters, 14B active per token, a 1M-token context window and over 200 tokens/sec generation. It reached 1668 points on the Code Arena WebDev leaderboard, edging past DeepSeek-V4.1-Flash and Kimi K3 and matching Qwen3.8-Max, with weights to be open-sourced under Apache 2.0 by November 25.

Arena.ai (@arena)T1·Models
78 pts
Baidu Unveils ERNIE 5.5 Flash Preview with 280B Parameters and 1M Context

Baidu released ERNIE 5.5 Flash Preview, built on a new architecture optimized for efficient inference, with 280B total parameters and 14B active per token, a 1M-token context window, and generation speed above 200 tokens/sec. The model is now live on Arena for testing, with Baidu claiming competitive results on Terminal-Bench and AutomationBench; open weights are slated for release by November 25.

IT之家 · 智能时代T2·Models
78 pts
Baidu Releases ERNIE 5.5 Flash Preview with 280B Parameters and Open-Weight Pledge

Baidu has released ERNIE 5.5 Flash Preview, the first post-trained variant of its Base model, featuring 280B total parameters, 14B activated parameters and a 1M-token context window, targeting coding, 3D modeling and content creation. The API is live on the Qianfan platform, with weights promised to be open-sourced by November 25; Baidu describes it as an intermediate post-training checkpoint still under rapid iteration.

Marktechpost (@Marktechpost)T2·Models
78 pts
Thinking Machines Ships Inkling 2 Small, a 594B Sparse MoE Model

Thinking Machines Lab released Inkling 2 Small, a 594B-parameter Mixture-of-Experts model that activates only 17B parameters per token, routing each token to 8 of 384 experts across 48 layers with a 292M vision encoder for image input. It scores 31 on the Artificial Analysis Intelligence Index and 86.3% on SWE-Bench Pro v2, but is slow per task and weak on Terminal-Bench at 13.1%; weights remain closed with an Apache 2.0 release planned after safety testing.

量子位 (微信公众号)T2·Models
40 pts
China Merchants' Lion Mountain Lab Releases 35B Embodied Reasoning Model with 3B Active Params

The Lion Mountain AI Lab under China Merchants Advanced Technology Institute has released Lion-ER-35B-A3B, an embodied-reasoning vision-language model built on a mixture-of-experts architecture with roughly 35B total parameters and only about 3B activated per token. The lab claims across-the-board gains on ten embodied-AI benchmarks, with some capabilities surpassing larger models, and attributes this to a capability-first data engine that decomposes embodied reasoning into 8 top-level domains and 39 sub-capabilities and iterates data against evaluation gaps rather than scaling raw data volume. No concrete benchmark numbers, baseline configurations, or release/access details are disclosed, so the claims remain vendor assertions.

新智元 (微信公众号)T2·Open Source
78 pts
Baidu Open-Sources ERNIE Unlimited OCR 2, Tops Global Long-Document Benchmarks

Baidu has open-sourced the next-generation lightweight document parsing model, ERNIE Unlimited OCR 2. With 3B total parameters but only 500M active during inference, it ranks first on major benchmarks like MarathonDocBench and MPDocBench. The key breakthrough is its ability to parse ultra-long documents continuously without splitting, demonstrated by processing thousands of PDF pages in a single pass while maintaining context, significantly lowering the barrier and cost for long-document AI processing.

腾讯混元 (Tencent Hunyuan)T1·Models
78 pts
Tencent Unveils Hy4 Preview Flagship Model with 770B Parameters and 1M Context

Tencent released Hy4 preview, a new flagship model with 770B total parameters, 49B activated parameters and a 1M-token context window, positioning it in the top tier of open models. Built with high-quality data co-created with Tencent's in-house software engineering, gaming, finance and security experts, it targets real productivity tasks such as long-horizon software development, office work and science.