最新模型

新模型、開放權重、多模態、語音、影像與模型架構。

GRSD Assigns Rewards by Comparing an Agent’s Own Success and Failure Trajectories; ALFWorld Unseen Task Success Rate Reaches 83.6%

Group-Reflective Self-Distillation no longer extracts fixed skills from an external teacher. Instead, it compares successful and failed rollouts from the same policy and converts their differences into turn-level training signals. It delivers overall improvements across three agent environments, but increases training time by 17% over GRPO and still requires an external model to score reflection text.

2026-08-01

GitHub Repo

GCC Rejects Copyright-Significant LLM-Derived Content, While Maintainers May Make Exceptions for Generated Tests

The GCC Steering Committee has approved an AI contribution policy that generally rejects substantial code and text containing or derived from LLM output. Models may still be used for debugging, analysis, and review, and maintainers may accept generated tests, but their output cannot be incorporated directly into ordinary contributions.

2026-08-01

代理工具

MemHarness Rewrites Agent Memory Before Acting, Raising 7B Model’s ALFWorld Success Rate to 85.2%

Instead of replaying retrieved past experiences verbatim, MemHarness has the same policy model critique and reorganize them based on the current state before deciding how to act. After end-to-end training with GRPO, it achieved success rates of 85.2% on ALFWorld and 75.6% on WebShop, though the code and model weights have not yet been released.

2026-08-01

代理工具

SciDataSailor Lets Agents Explore Scientific Databases Directly, Doubling the 9B Model’s Pass@1 from 14.01 to 28.99

SciDataSailor uses Monte Carlo tree search to synthesize executable file-exploration trajectories and has released more than 2,300 supervised examples. After fine-tuning, Qwen3.5-9B made fewer trial-and-error attempts in evaluations with constrained step budgets, though some scoring still relies on an LLM judge.

2026-08-01

代理工具

RepoComplianceBench: Four Coding Agents Almost Never Proactively Read Open-Source Projects’ AI Contribution Rules

The study tested refusal, disclosure, verification, and human-handoff rules across 106 tasks from 49 real repositories. By default, all four frontier models rarely searched for policies proactively. Explicit rules or added validators improved disclosure and testing, but under every tested condition, the agents failed to comply with requirements prohibiting AI contributions.

2026-08-01

AI 安全

Microsoft Routes Vulnerability Analysis Through a Security Model With 5B Active Parameters; MDASH Scores 95.95% on CyberGym

MAI-Cyber-1-Flash is a sparse MoE model with 137B total parameters, but activates only about 5B per token. It handles roughly 90% of MDASH’s workload before passing difficult cases to GPT-5.4. Microsoft says the new configuration cuts costs by nearly half, but the model is currently available only in private preview on Azure AI Foundry, and the benchmark results have yet to be independently reproduced.

2026-07-29

AI 安全

InfoOps Bench Updates Its State Propaganda Test Set Weekly; Refusal Rates Across 17 Models Differ by 85.7 Points

InfoOps Bench continuously draws narratives from Russian, Chinese, and Iranian state-backed media to test whether models refuse, fact-check, or help rewrite propaganda content. Integrity scores across 17 models range from 8.8% to 94.5%, showing that model size is insufficient to predict information-operations risk.

2026-08-01

推論系統

Local LLM Testing on M4 Max: Faster Decoding Than GB10, but Full Responses Remain Constrained by Prefill

A new round of similarly priced local AI testing shows that the M4 Max, with 546GB/s of unified memory bandwidth, leads the GB10 and Strix Halo in per-token decoding across three quantized models. The GB10 still has the advantage in prompt prefill and overall energy efficiency, demonstrating that tokens/s alone can misrepresent actual response times.

2026-08-01

SVR Lets a 2B Model Decide When to Stop Revising, Completing Mathematical Reasoning in 2.99 Rounds on Average

SVR jointly trains answer generation, correctness judgments, and confidence estimation, allowing the model to decide at inference time whether to retain an answer or continue refining it. Qwen3.5-2B achieved 56.3% average accuracy across seven mathematics benchmarks, although its stopping signal may still be affected by the model’s own calibration biases.

2026-07-31

開發者平台／API

GitHub Models Officially Shuts Down, Taking Its Inference API, Model Catalog, and BYOK Offline

GitHub shut down the Models playground, model catalog, inference API, and bring-your-own-key functionality on July 30, including for existing paying users. Applications that depend on its endpoints must migrate to Microsoft Foundry, another model gateway, or an in-house provider abstraction layer.

2026-07-23

AI 安全

Seven Simple Image Transformations Can Bypass Three Commercial AI Content Moderation Services

Black-box experiments found that low-cost transformations such as color inversion and grayscale conversion—which require neither gradients nor surrogate models—can cause three commercial services to reclassify still-recognizable harmful images as safe. Multimodal content and the self-harm category were especially vulnerable, showing that foundation model–based moderation APIs should not be treated as a single safety boundary.

2026-07-31

AI 安全

MIND Filters Agent Memory with an Intent Information Bottleneck, Cutting Attack Success Rates by More Than 55% on Average

Rather than invoking a large model to review each memory individually, MIND compares compressed representations of the original intent and subsequent behavior, then uses a multi-hyperplane classifier to intercept poisoned content. Controlled experiments with four models showed lower attack success rates and average latency close to the undefended baseline, although the code has not yet been released.

2026-07-31