Archive
Agent、MCP、coding agent、工作流自動化與評測。
代理工具
Cohere Labs has released the Agentic Task Ecosystem, which aggregates 696,291 tools from seven directories and maps them to tasks in the U.S. O*NET database. The results suggest that many MCP tools cover only fragments of a workflow, but the 2.6% figure comes from semantic matching and model-based classification—not real-world execution tests.
最新模型
Ant Group’s inclusionAI has released LLaDA-Image weights and Diffusers inference code. A single model handles text-to-image generation, reference-image editing, and Chinese and English text rendering. The distilled version uses Twin-DMD to reduce the recommended sampling process from 50 steps to 4, but training code and independent performance validation are still unavailable.
代理工具
The Agent Merge preview extends coding agents’ scope from generating changes to finalizing PRs, allowing them to repeatedly address review feedback, failed checks, and branch conflicts. Once enabled, the session switches to Autopilot and Assisted permissions, so teams must treat automatic merging and required checks as a privileged control plane.
推論系統
The new open-source inference server Magnitude benchmarks hardware before recommending a complete combination of model, quantization, context length, and estimated speed. It automates the memory and concurrency settings most often misconfigured for local agents, though performance and model quality currently still depend largely on the project’s catalog and testing.
AI 平台與基礎設施
NVIDIA has agreed to acquire Hugging Face, gaining control of a key distribution hub for models, datasets, applications, and inference services. The company has pledged that the platform will not be tied to NVIDIA hardware, but its governance, ranking practices, and cross-hardware optimization after the deal will require continued scrutiny.
最新模型
OpenAI is making Astra available through the API, AWS, and paid ChatGPT plans, while experimenting in Codex with notes and history search that span context windows. The model’s agentic and cybersecurity capabilities have improved significantly, but its chain of thought is better at evading monitoring, and deployment safeguards may also terminate legitimate work.
最新模型
Puffin-World uses a nine-channel Omni-Camera condition to unify absolute physical orientation with relative camera motion, while handling understanding, generation, and 3D reconstruction within a single architecture. The code, weights, and Puffin-16M dataset have been released, but its physical state modeling is currently focused mainly on gravity and latitude in static scenes.
代理工具
Hugging Face has open-sourced Funes, which writes turn-by-turn logs from Claude Code, Codex, pi, and Hermes into a single Lance dataset. Retrieval and reranking run locally by default, but cross-machine synchronization, residual secrets, and prompt injection through external memory still require user-managed safeguards.
最新模型
Hcompany has open-sourced the 260M- and 800M-parameter NeoMME models, which process text tokens and raw image patches with the same encoder instead of attaching a vision tower or causal decoder. Its retrieval variants produce both dense and late-interaction representations, although the current efficiency and accuracy figures come primarily from evaluations conducted by the authors.
最新模型
Microsoft’s new speech recognition model supports 60 languages and natively provides speaker diarization, terminology biasing, code-switching, and word-level timestamps. Independent API testing shows strong error rates and batch throughput, but its promotional pricing, Chinese-language performance breakdown, and streaming latency remain to be verified.
推論系統
PAIR adds a compatible proxy layer between Ollama, LM Studio, and agents, routing independent requests to computers on the local network according to model availability and GPU load. It increases workload concurrency; it does not pool VRAM or split a single inference across multiple GPUs.
AI 安全
Exclusion rules configured by Copilot Business and Enterprise administrators can now prevent the Copilot App and CLI from directly adding specified files to model context. The mechanism is not comprehensive data loss prevention or an execution sandbox; GitHub still lists gaps involving symbolic links, remote file systems, and indirect IDE semantics.