推論系統

推論 runtime、量化、 serving、KV cache、成本與效能。

MissClick Exploits Coordinate Token Place Values to Attack GUI Agents, Achieving Targeted Misclick Rates of Up to 62.67%

Rather than treating GUI coordinates as ordinary text, MissClick exploits small changes to hundreds-place tokens to cause large shifts in click positions. The study significantly increased attack success rates against OS-Atlas and UGround, showing that the security of computer-use agents cannot be evaluated solely by element-localization accuracy.

2026-08-05

最新模型

ReflectRL Turns Expert Error Trajectories into Training Signals, Improving GRPO Cross-Domain Evaluation by an Average of 19.1 Points

Instead of discarding reasoning trajectories where a strong model produces an incorrect answer, ReflectRL asks the student model to identify localized errors and then gradually removes this hint. The study reports consistent gains across nine benchmarks, although the results come primarily from mathematical reasoning settings, and the listed GitHub project link is currently inaccessible.

2026-08-05

代理工具

GitHub Spark Stops Creating New Apps; Existing Projects Using llm() Must Replace Their Inference Service

GitHub has stopped accepting new Spark users and allowing new apps to be created. Existing users have until August 31 to export their code. Deployed sites can continue operating, but llm() calls that depend on GitHub Models have stopped working, requiring developers to integrate an external model provider and take over API key and billing management.

2026-08-05

AI 安全

Science Reasoning Benchmark Shows “Right Answer, Wrong Method,” With Shortcut Solutions Reaching 37.4% on HLE

A new study finds that models can obtain correct results without completing the required derivation by using shortcuts such as guessing formulas, enumeration, or guessing an answer first and then verifying it. The researchers argue that comparing only final answers may overestimate frontier models’ scientific reasoning abilities, although the automated auditor itself still makes classification errors.

2026-08-05

最新模型

Faster-WAM Shrinks the Robot Action Head to a Single Layer, Cutting Control Inference Latency to 66.5 Milliseconds

Faster-WAM attaches a single-layer action head to a 30-layer video Transformer and obtains world-model representations through cross-layer KV fusion, eliminating the need for the action branch to replicate the full backbone depth. The study reports a 3.2× speedup over Fast-WAM, but code and weights for the new architecture have not yet been released, and the results come primarily from simulation benchmarks.

2026-08-04

代理工具

Harness-R1 Learns to Modify Execution Frameworks from Agent Failure Trajectories, Boosting Qwen3.5-9B Success Rate by 9.3 Points

Rather than adjusting the target agent’s model weights, Harness-R1 trains a separate 9B model to modify its context, tools, and validation code. The average success rate across three environments rose from 44.3% to 53.6%, but no public code is available yet to assess patch safety.

2026-08-04

最新模型

CMuon Splits Fused DiT Weights Before Orthogonalization, Training a 675M Model Twice as Fast as AdamW

CMuon finds that applying Muon orthogonalization to fused QKV, AdaLN, and FFN weights as a whole causes interference between functionally distinct subspaces. The study improves late-stage convergence through blockwise momentum orthogonalization, although the claimed 2× speedup has so far been validated only on a single 675M DiT configuration.

2026-08-04

推論系統

xPress Restores Causal Dependencies to Diffusion-Based Drafts, Boosting Qwen3-8B Decoding Throughput by About 1.3× on Average

xPress adds a single parallel causal refinement step between block-diffusion drafts and the target LLM, reducing early rejections caused by drafts in which individual tokens are plausible but the sequence as a whole is not. The study reports an average increase of about 30% in accepted draft length, but no public implementation is currently available for independent reproduction.

2026-08-04

代理工具

AgentTrajectorySentinel Uses Microsecond-Level Temporal Monitoring to Stop Agents From Going Off the Rails, Raising Task Success Rate From 52% to 73% in Tests

A new study replaces step-by-step review by a second LLM with trajectory-drift detection based on telemetry such as tool actions, latency, outputs, and token probabilities. The system costs about 200 microseconds per step, but each deployment must recalibrate it using its own normal trajectories.

2026-08-04

代理工具

GitHub Copilot Cloud Agent Adds Reasoning Effort Selection, Letting Complex Tasks Use More AI Credits for Deeper Analysis

GitHub now allows developers to specify both the model and reasoning effort when launching Copilot cloud agent tasks. Higher effort may improve cross-file debugging and architectural analysis, but it increases token usage and AI Credits consumption, requiring teams to establish new cost and quality benchmarks.

2026-08-04

推論系統

Cloudflare Separates Prefill and Decode Precision, Boosting GLM 5.2 Decode Throughput by Up to 55%

Cloudflare uses an FP8 KV cache to expand Kimi K2.6’s concurrency capacity and switches to INT4 weights for GLM 5.2 decoding. The system also adds page-generation tags to the shared KV cache to prevent multi-tenant requests from reading incorrect cached data.

2026-08-04

最新模型

MiniMax H3 Opens Video Generation Weights, ComfyUI Simultaneously Provides a Runnable Workflow

MiniMax has released the weights for its H3 video model, enabling developers to run image-to-video generation with synchronized audio locally. ComfyUI has added an official workflow, but consumer GPU tests largely rely on pruned and INT8 derivative files and should not be treated as full-quality reproductions of the original weights.

2026-08-04