#Model releases

Tencent Mixuan releases WeVisOne-VL-4B unified visual language model

Tencent Mixuan released the WeVisOne-VL-4B model, built on Qwen3-VL, integrating a 9.6M parameter pixel decoder, supporting object-level perception (open-vocabulary object detection, pointing localization, counting, pointing and OCR) and pixel-level perception (pointing segmentation and promptable alpha friction), suitable for research tasks such as visual localization, pointing, counting, OCR localization, pointing segmentation, and image friction.

Meta Expert: Cross-layer meta expert sharing for implementing efficient MoE models

Meta Expert Model Reduces Redundancy Through Cross-Layer Meta-Expert Sharing, Compared to Layer-Local Expert MoE Models, It Has Lower Parameter Redundancy, Lower Cross-Entropy Loss, and Higher Accuracy on Downstream Tasks. When Activation Amount and Expert Granularity Are Fixed, This Model Achieves Performance Comparable to Vanilla MoE Using Only 70-80% of Experts, Significantly Enhancing Model Scalability. Theoretical Analysis Shows that Meta-Experts Provide More Flexible Residual Paths, Thus Improving Information Retrieval Efficiency.

Deep Fusion Memory: A block-level parameterized memory architecture for large language models

Deep Fusion Memory (DFM) integrates predicted parametric memory into intermediate hidden states through gated cross-attention, reducing retrieval overhead and improving the performance of knowledge-intensive tasks without the need for online retrieval.

Reports say memory manufacturers have started preliminary work for the SOCAMM3 module

PCB manufacturer TLB has held early discussions with two memory manufacturer customers on the design and manufacturing plan for the SOCAMM3 memory module, but the final specifications have not been determined. The mass production of SOCAMM3 is expected to start in 2028 and may enter the market along with NVIDIA's Rosa CPU.

vLLM provides zero-day support for MiniMax-M3.1-Flash

vLLM provides zero-day support for MiniMax-M3.1-Flash on NVIDIA Blackwell, Rubin, and AMD MI350 series GPUs, performing excellently on agent workloads.

Cola2: Adaptive Hierarchical Language Model improves hierarchical language generation through continuous semantic compression and balanced discrete generation.

Cola2, through dynamic span splitting, masked diffusion decoding, and stream matching priors jointly optimizes continuous semantic compression, reconstruction-generation balance, and encoder-decoder adaptation, verifying its core design on five downstream tasks. Using 2100 billion training tokens, it demonstrates competitive non-saturated scalability, with a 3460 billion source token checkpoint showing an average of 60.10 points on eight tasks and surpassing Llama-3-8B on HellaSwag and StoryCloze.

OpenAI widely opens its decision-making model Decisions API, intensifying competition in the decision-making model market

OpenAI has widely opened its Decision API to users for the first time, a model that reconstructs ChatGPT Luna to achieve faster inference and lower costs, reducing speed from second-level to millisecond-level.

Boundless Dynamics has released and open-sourced the MWA™-Phoenix Subspace World Model Foundation.

Boundless Dynamics has released and open-sourced the MWA™-Phoenix Subspace World Model Foundation, sweeping the LIBERO-Plus, LIBERO-Pro, RoboCasa GR1, Bench2Dex, RoboTwin 2.0, and RoboDojo real machine rankings.

Hugging Face releases IBM Granite time series model integration and prediction solution

Hugging Face announced IBM Granite, a time series model integration and forecasting solution, combining four Granite models (PatchTST-FM-r1, PatchTST-FM-r2, FlowState-r1.1, TTM-r3) to provide out-of-the-box integrated forecasting.

Tencent Mixuan open-sourced the real-time simultaneous translation model Hy-SimulMT-30B-A3B

Tencent Mixuan has released and open-sourced the real-time simultaneous translation model Hy-SimulMT-30B-A3B, supporting bidirectional Chinese-English voice-to-voice translation, capable of continuously outputting translated text and speech during speech, and supporting real-time voice cloning.

Cactus Whistle 16.9 MB Speech Model Performance Evaluation on CPU: Comparison Results with Whisper Base

The Cactus Whistle speech model transcribes seven languages on the CPU with a 16.9 MB file size and returns word timestamps and probabilities. Tests show it outperforms Whisper Base on its test-other dataset, but there is no significant difference from beam search, with the speed advantage shrinking as audio length increases. Additionally, it tested 30-second window audio processing, streaming latency, timestamps, keyword bias, other languages, noise, embedding, and speech-to-tool calling features.

Nsight Python Tutorial: How to analyze GPU kernel performance

Use Nsight Python (NVIDIA's open-source Python interface) to analyze GPU kernel performance, mark code with `nsight.annotate`, decorate functions with `nsight.analyze.kernel` to obtain a DataFrame of hardware metrics for each code segment.

Recursive releases the open-source 40B fuzzy model Flint, designed specifically for agents.

Recursive releases the open-source 40B fuzzy model Flint, 2.5B active parameters, performs best among MoE models below 50B, supports inference and OpenAI compatible tool calls, 4-bit weights about 23GB, SWE-bench Verified 74.2%.

TorchTitan update: new SOTA model support, scaled performance improvement, and TitanRL release

TorchTitan adds native PyTorch support for the latest SOTA models including Muse Spark Open, Muse Glimmer, Qwen3-VL, Qwen3.5, Kimi K2.7, Kimi K3, and DeepSeek-V4.

How Inception Mercury 3 beats GPT-6 Luna and Claude Haiku 5.5 with less wait time.