📰 Dev.to · pueding

Cluster-Route-Escalate Cascade Serves LLMs at 97-99% Accuracy for Less Cost: Cost-Aware LLM Cascade

S-Agent: Spatial Tool-Use Makes an 8B Agent Rival GPT-5.4 on Spatial Reasoning: Spatio-Temporal Evidence Accumulation

SGLang v0.5.14: LPLB Expert-Parallel Load Balancing

OpenAI and Broadcom's Jalapeño, a Custom Inference ASIC: Inference ASIC vs GPU

Baidu Unlimited OCR Holds the KV Cache Constant for 40+ Pages: Reference Sliding Window Attention

GLM-5.2 Becomes the Top Open-Weights Model: Active vs Total Parameters

Agent Leaderboards Mislead Under Distribution Shift (IBM): Predictive Validity

AMD ATOM + ATOMesh: Prefill/decode Disaggregation on ROCm

NVIDIA Blackwell Sweeps MLPerf Training 6.0: Strong Scaling

NVIDIA Blackwell Leads AgentPerf, the First Agentic-AI Infra Benchmark: Trajectory-Replay Benchmarking

NVIDIA RTX Spark Superchip: Unified CPU–GPU Memory

Google Ships Gemma 4 QAT Checkpoints: Quantization-Aware Training

MiniMax M3 Ships Open-Weight 1M Context: MiniMax Sparse Attention (MSA)

Google Releases DiffusionGemma: Parallel Block Decoding

AutoLab Benchmarks Frontier Agents on Long-Horizon R&D Tasks: Iterative Experiment-Loop Evaluation

AgentDoG 1.5: Small Inline Guard Models for Agent Actions

Claude Opus 4.8: Parallel-Subagent Dynamic Workflows

OmniRetrieval: Source-Native Query Dispatch

Gemini 3.5 Flash: Agent-First Model Design

Cursor Composer 2.5: Targeted Textual Feedback RL

Boiling the Frog Paper: Multi-Turn Norm Erosion vs Single-Prompt Agent Safety

OpenSCAD Pantheon Benchmark: Human-In-The-Loop vs Autonomous Coding Agents

Camouflage Injection Paper: Camouflage Detection Gap

MCP SEP-2468: RFC 9207 Iss Parameter for OAuth Mix-Up Defense