硬體

AI 晶片、GPU、資料中心、邊緣裝置與硬體安全。

NemoClaw 0.0.114 Moves Read-Only MCP Calls Out of the Model Decision Path and Tightens Image and Installation Verification

NVIDIA has added deterministic read-only MCP calls to LangChain Deep Agents Code, allowing automated workflows to execute restricted tools without model involvement. The new release also pins container image digests, strengthens supply chain verification, and improves sensitive export cleanup, although “read-only” still depends on the accuracy of tool registration metadata.

2026-08-24

最新模型

Wan3.0 Officially Launches: Generate 30-Second Videos with Audio in a Single Run, Using Documents and Web Pages as Direct Inputs

Alibaba has officially launched Wan3.0 following its invite-only testing in early August, unifying support for text, images, video, audio, documents, and web-based references. The API is now public, but the model remains labeled invite-only, with no weights, architecture details, or independent evaluations released.

2026-08-24

推論系統

llama.cpp 0.2.0 Separates Stable Releases from Nightly Builds, Finally Letting Package Maintainers Pin Compatible Versions

llama.cpp 0.2.0 officially introduces semantic versioning for slower-moving stable releases, while retaining the existing `b[number]` tags for development builds produced for nearly every commit. The new system reduces versioning risks for Debian, Homebrew, and downstream bindings, but “stable” does not mean long-term support or comprehensive validation across hardware platforms.

2026-08-24

最新模型

GPT-5.6 Sol Temporarily Drops to $20 per Million Output Tokens, While Entire Requests With Ultra-Long Prompts Still Incur Higher Rates

OpenAI has cut GPT-5.6 Sol API input and output prices by 20% and 33%, respectively, with the promotional rates lasting at least through November 21. Once input exceeds 272K tokens, the higher rates still apply to the entire request, so long-context agent costs cannot be estimated from standard rates alone.

2026-08-24

推論系統

FreeToken Dynamically Allocates CPU, GPU, and PCIe Resources to Run a 35B MoE Model with 8GB of VRAM

Instead of using fixed offloading, FreeToken uses measured bandwidth to decide whether to transfer cache-missed experts to the GPU or execute them directly on the CPU. The paper also demonstrates a 753B model served by a single workstation GPU, but the full weights still require system memory; this should not be interpreted as “fitting all 753B parameters on one GPU.”

2026-08-23

代理工具

NVIDIA AVO Solves All 183 Levels in the ARC-AGI-3 Public Set, With Key Gains Coming From the Agent Framework Rather Than a Model Change

Using persistent memory, a supervisor agent, and a text-grid interface, AVO enabled Claude Opus 5 to complete every public level within 6,624 environment actions. The score of 100 covers only the public set and should not be interpreted as passing the semi-private or private competition tests.

2026-08-23

最新模型

IAR Splits Document Internalization Into Injection, Question-Answer Alignment, and Weight Merging, Preserving More General Capabilities on Average

A Beijing Academy of Artificial Intelligence research team has proposed a three-stage post-training method that enables models to answer questions about a fixed document set without RAG. IAR outperformed standard SFT across all metrics in seven of eight experiments, although the results still depend on synthetic question-answer pairs, model judges, and model-specific hyperparameter tuning.

2026-08-23

最新模型

DeepSeek V4-Flash-Vision-Exp Adds Image Input and a 1M-Token Context Window, but Remains an Experimental API

DeepSeek has added image understanding to V4-Flash, supporting text and images through three compatible interfaces while retaining the existing Flash pricing. However, its officially reported agent benchmark results have yet to be independently reproduced, and neither the model weights nor the vision architecture has been disclosed.

2026-08-23

推論系統

SGLang 0.5.18 Overlaps Weight Loading with CUDA Graph Capture, Cutting Qwen3-32B Cold Start to 35.6 Seconds

The new release processes checkpoint staging and CUDA Graph capture concurrently during startup, delivering a 2.38× speedup over the standard default workflow in official H100 testing. It also rewrites parts of the tensor-parallel communication path and adds seven model families, but the Torch upgrade and cache-path consolidation introduce migration costs.

2026-08-23

硬體

Needle 2 Compresses a 45M-Parameter Tool Model into 14MB and Constrains On-Device Calls with Schema Grammars

Cactus Compute recently improved Needle 2’s Python API, cross-platform execution, and Apache 2.0 licensing, while continuing to update its model card. It performs tool selection and structured extraction with 28MB of peak memory, although its speed and accuracy claims still come primarily from official tests.

2026-08-23

推薦系統

NVIDIA Rewrites KV and Beam Search for Its Generative Recommendation Stack, Making 5K-Context Inference 2.27× Faster

NVIDIA has assembled HSTU, dynamic embeddings, and Semantic ID-based generative recommendation into an executable PyTorch reference stack. It rewrites the caching and decoding paths for recommendation workloads characterized by long histories, generation of only a few tokens, and extremely large beam widths, though the results are currently limited to NVIDIA hardware and synthetic configurations.

2026-08-22

推論系統

LFM2.5 Adds DSpark Draft Models, Boosting H100 Decoding Throughput by Up to 3.18×

Liquid AI has released DSpark draft checkpoints with about 300 million parameters for three LFM2.5 models and integrated them with llama.cpp and SGLang. Official tests show that acceleration varies significantly by model, data distribution, and hardware, so peak results should not be treated as representative of gains across all workloads.

2026-08-22