Archive
新模型、開放權重、多模態、語音、影像與模型架構。
代理工具
Anthropic has increased Claude Code’s default subagent nesting depth from one level to three and incorporated deeper agents’ output into structured streaming. The new version can also immediately reject network hosts that are not on the allowlist, making it easier for unattended agents to adopt a default-deny policy.
AI 基礎設施
Hetzner has quietly opened an experimental LLM inference service that currently offers only Qwen3.6-35B-A3B-FP8. Existing OpenAI SDKs can connect by changing the base URL. Low-load community tests recorded a median time to first token of 153 milliseconds and roughly 224 tokens/s, but the service has no SLA, billing, or production guarantees.
推論系統
An inference-time experiment on an unreleased 27B hybrid model found that it retrieved only 21.4% of injected information at roughly 1 million tokens, far below the 92.9% achieved at 33,000 tokens. Position scaling, attention temperature, and a higher-precision KV cache failed to significantly fix the problem, while retrieval capability was highly concentrated in 18 attention heads.
AI 安全
QuantiBias holds the model and prompts constant while varying only weight precision, measuring multilingual open-ended bias, refusals, and multiple-choice performance before and after quantization. The authors find that common short-form safety metrics remain broadly stable, yet an independent judge model still classifies about 24% to 27% of open-ended responses as endorsing stereotypes.
推論系統
Hugging Face Diffusers can now directly load Nunchaku Lite checkpoints to run diffusion Transformers with 4-bit weights and activations. Official single-machine tests showed up to roughly half the peak VRAM usage and a 1.8× speedup, but hardware coverage and image-quality validation remain limited.
代理工具
GitHub MCP Server now supports the MCP `2026-07-28` specification, scheduled to be finalized on July 28. Remote requests no longer depend on the `initialize` handshake or protocol-layer sessions. The new design simplifies horizontal scaling, but compatibility work remains for Tasks, error codes, and custom clients.
代理工具
Open Code Review delegates file filtering, rule matching, and comment placement to deterministic code, then lets LLM agents search repository context and identify defects. Alibaba’s internal benchmarks show token usage at roughly one-ninth that of Claude Code, but it trades lower recall for higher precision and cannot directly replace existing tests and static analysis.
最新模型
Microsoft is bringing its highest-quality image generation model and a high-throughput voice model into public preview in Foundry simultaneously, targeting precision editing and real-time voice agents, respectively. The company has published cost and latency data from several production environments, but cross-provider comparisons still lack consistent, reproducible testing conditions.
最新模型
SGLang-Omni has added a deployment path for MOSS-Transcribe-Diarize that returns segment-level timestamps, text, and speaker labels in a single response through an OpenAI-compatible endpoint. This makes the open 0.9B model easier to deploy in online services, although the three evaluation datasets used for the official performance results are not currently fully public.
最新模型
Upstage has released the weights for Solar Open 2, which uses 250B total parameters and activates 15B parameters per token for long-horizon agentic workflows. The model retains softmax attention and a KV cache in only one-quarter of its layers, but its million-token capability and Korean office-agent results are still based largely on the team’s own evaluations.
代理工具
Meta has deployed Muse Spark 1.1’s planning and tool-use capabilities to Meta AI, enabling the assistant to read email and calendars and create persistent outputs such as presentations. The update expands the private context accessible to the model, but the company has not disclosed tool permissions, revocation mechanisms, or end-to-end task success rates.
AI 基礎設施
NVIDIA has detailed ModelExpress’s complete transfer path, allowing new inference replicas to copy weights and compilation caches from existing GPUs, with P2P RDMA as the preferred method. In an eight-B200 test, the company cut DeepSeek-V4-Pro’s time to API readiness from eight minutes to 1 minute 44 seconds, though the result depends heavily on identical GPU models, high-speed networking, and compatible tensor configurations.