1
银行理财“信披明牌”不止指数基准，有机构尝试展现理财产品最大回撤
当前，多家理财公司正将理财产品的业绩比较基准切换为“动态指数型”。某头部股份行理财公司产品部人士向记者表示：“公司正全面梳理产品信息。以后业绩比较基准将无法随意调整，所以每只产品都要根据数据测算，挑选与产品定位和长期走势相吻合的指数组合作为业绩比较基准。”面对“换锚”后的投资者理解难题，理财机构及商…
2
FIRE-Bench: Evaluating AI Agents on the Rediscovery of Scientific Insights
一个评估AI智能体能否“重新发现”科学洞察的创新基准，ICML 2026收录，值得研究者关注。
arXiv:2602.02905v2 Announce Type: replace Abstract: Autonomous agents powered by large language models (LLMs) promise to accelerate scientific discove…
3
Imaging-101: Benchmarking LLM Coding Agents on Scientific Computational Imaging
新基准Imaging-101系统评估LLM编码代理在科学计算成像上的表现，揭示当前模型能力边界。
arXiv:2607.10789v1 Announce Type: new Abstract: Computational imaging, which recovers hidden signals from indirect, noisy measurements, underpins quan…
4
Capabilities of Claude Fable 5 on Biomedical Challenge Problems
首个公开评估Claude Fable 5在生物医学挑战问题的表现，直面基准饱和与评估方法可靠性问题。
arXiv:2607.10849v1 Announce Type: new Abstract: Frontier language models are increasingly evaluated on biomedical benchmarks, but two problems undermi…
5
VoxENES 2026: Benchmarking Generalization of Speech Spoofing Detectors Against LLM-Era TTS and Voice Conversion
LLM时代语音合成与转换加剧深度伪造风险，这项新基准测试揭露了现有欺骗检测器的泛化短板。
arXiv:2607.11706v1 Announce Type: cross Abstract: Modern LLM-driven text-to-speech (TTS) and voice conversion (VC) systems produce synthetic speech th…
6
Do Video-LLMs Actually Watch? Diagnosing Character-Tracking Failures in Long-Form Video
诊断Video-LLM在长视频中跟踪角色的失败，揭示当前最强开源模型未能真正“观看”视频的局限。
arXiv:2607.11078v1 Announce Type: cross Abstract: Can a Video Large Language Model (Video-LLM) follow one person through a long video, keeping track o…
7
RVN-Bench: A Benchmark for Reactive Visual Navigation
全新基准测试RVN-Bench，专为反应式视觉导航设计，推动机器人自主导航评估标准化
arXiv:2603.03953v2 Announce Type: replace-cross Abstract: Safe visual navigation is critical for indoor mobile robots operating in cluttered environme…
8
SLVMBench: Skill Learning from Video Memory
新基准测试SLVMBench，评估AI从视频记忆中习得技能的潜力。
arXiv:2607.11312v1 Announce Type: new Abstract: We introduce Skill Learning from Video Memory (SLVMBench), the first benchmark that jointly evaluates …
9
How to Review an Edge LLM Benchmark Without Fooling Yourself
从Jetson Nano到模型变体，揭示边缘LLM基准测试中易忽视的变量，避免自欺欺人。
A Jetson Nano and Ollama benchmark review appeared among DEV's top articles on 2026-07-12 . Edge inference experiments are valuable because they make …
10
IMProofBench: Benchmarking AI on Research-Level Mathematical Proof Generation
60位作者联合发布，首个评估AI在研究级数学证明生成能力的基准测试，挑战大模型推理极限。
arXiv:2509.26076v2 Announce Type: replace Abstract: As the mathematical capabilities of large language models (LLMs) improve, it becomes increasingly …
11
DeepSWE: Measuring Frontier Coding Agents on Original, Long-Horizon Engineering Tasks
新基准DeepSWE用113个原创、长周期软件工程任务评估编码代理，避免SWE-bench的数据泄露问题。
arXiv:2607.07946v1 Announce Type: cross Abstract: DeepSWE is a benchmark of 113 original, long-horizon software engineering tasks for evaluating codin…
12
AI Chatbot Suicide Risk Detection and Response: Human Validation Study of the Open-Source VERA-MH Safety Evaluation
开源基准测试工具，专为评估AI聊天机器人在心理健康场景中的安全性，尤其聚焦自杀风险检测与响应能力。
arXiv:2602.05088v4 Announce Type: replace Abstract: Millions of people now use generative AI chatbots for psychological support. Despite their promise…
13
Does AI Understand Imaging? A Systematic Benchmark of Agentic AI for Computational Imaging Tasks
系统化基准测试揭示智能体AI在计算成像任务中的真实表现与理解局限
arXiv:2607.07189v1 Announce Type: new Abstract: Vision-language models (VLMs) and agentic AI have shown strong performance on semantic visual tasks, b…
14
Agentic test processes, LLM benchmarks, and other notes on agentic coding fr
AI代理编程的荒诞体验：如果人类做错事会被解雇，但面对AI你却想放大这种错误到上千倍。
Article URL: https://danluu.com/ai-coding/#llm-variance Comments URL: https://news.ycombinator.com/item?id=48836951 Points: 6 # Comments: 1
15
Separating signal from noise in coding evaluations
OpenAI揭秘热门编码基准测试的缺陷，重新审视AI模型评估的可靠性
A new analysis from OpenAI reveals issues in SWE-Bench Pro, a popular coding benchmark, raising concerns about reliability and accuracy in evaluating …
16
OpenAI 挑战 AI 评测基准 SWE-Bench Pro：约 30% 存在评测缺陷
行业权威SWE-Bench被OpenAI揭短：约30%测试任务存在缺陷，8个月通过率飙升3倍多，AI编程能力评估面临重建。
IT之家 7 月 9 日消息，OpenAI 昨日（7 月 8 日）发布博文， 挑战行业权威评测基准 SWE-Bench Pro，认为在 731 个公开测试任务中，约 30% 存在评测缺陷。 IT之家注：SWE-Bench Pro 是由 Scale AI 推出的大语言模型与 AI 智能体编程能力评测基…
17
Google updates Android Bench with new LLMs, but Gemini still lags behind
Google更新Android基准测试，新增LLM评测，自家Gemini竟拖后腿了。
Android Bench is evolving, and developers can help guide that process.
18
马来西亚央行将基准利率维持在2.75%不变
7月9日，马来西亚央行将基准利率维持在2.75%不变。马来西亚央行称，强劲的基本面将继续支撑国内经济抵御外部冲击的韧性；预计2026年整体通胀将保持在可控水平，反映国内政策措施和稳定的需求状况。（界面）
19
HoloCount: A Holistic Visual Counting Benchmark for MLLMs
多模态大模型视觉计数能力新基准，HoloCount全面评测模型在复杂场景下的计数表现，揭示当前MLLMs的瓶颈。
arXiv:2607.06420v1 Announce Type: new Abstract: Visual counting is a fundamental pillar of multimodal intelligence, requiring a seamless integration o…
20
Constellation Gate AI
防提示注入+节省token，基准测试第一的AI安全工具，效率与安全兼得。
Prompt injection and token savings - #1 in benchmarks Discussion | Link