Ethan Mollick@emollickAI 评分7272We evaluated an early version of Claude Mythos Preview for risk assessment during a limited window in March 2026. We estimated a 50%-time-horizon of at least 16hrs (95% CI 8.5hrs to 55hrs) on our task suite, at the upper end of what we can measure without new tasks.
Baidu Inc.@Baidu_IncAI 评分2626DuMate的智能体能力刚经历测试，结果出炉：在@pinchbench和DeepResearch Bench上均位列第一。 详细分析如下👇
Artificial Analysis@ArtificialAnlysAI 评分6363http://x.com/i/article/2051782974098886656
Rohan Paul@rohanpaul_aiAI 评分6868http://x.com/i/article/2052027135619919876
Xiaomi MiMo@XiaomiMiMoAI 评分5959BREAKING: MiMo-V2.5 by @Xiaomi is #6 overall out of open weight models on Design Arena! This has an Elo of 1297 and is in the same performance band as Kimi K2.5 (Thinking) by @Kimi_Moonshot Congrats to the @XiaomiMiMo team on the launch!
Hugging Face 的 Open ASR Leaderboard 加入了来自 Appen Inc. 和 DataoceanAI 的高质量私有英语 ASR 数据集，涵盖多种口音及脚本式/会话式语音，总时长约 28.7 小时。数据集保持私有以降低 benchmaxxing（针对基准的过度优化）或测试集污染风险。默认平均 WER 仍基于公开数据集计算，用户可通过切换查看私有数据集影响。标准化采用基于 Whisper 的标准化器，UI 代码和评估脚本已开源。自 2023 年 9 月上线以来，该榜单访问量已超过 71 万次。
Jeff Dean@JeffDeanAI 评分5050📢Keynote announcement: @andykonwinski (Andy Konwinski), co-founder of Databricks and Perplexity AI, founder of @LaudeInstitute, is keynoting #CAIS2026. Laude's "Ship Your Research" mission funds open-source AI research through its Moonshots, Slingshots, and Open Frontier programs. Case in point: Terminal-Bench, a Laude-backed agent benchmark, reached Anthropic's Claude 4 model card 126 days after inception and has since become the industry-wide standard for measuring command line performance. We're thrilled to have him. San Jose · May 26–29 https://caisconf.org
Ethan Mollick@emollickAI 评分6666Rohan Paul@rohanpaul_aiAI 评分6868Artificial Analysis@ArtificialAnlysAI 评分5858
Luma@LumaLabsAIAI 评分7171Exciting news: UNI-1.1-Max and UNI-1.1 debuts making @LumaLabsAI the #3 lab in the Image Arena across both Text-to-Image and Image Edit! These are versions released without agentic search. Text-to-Image Arena - UNI-1.1-Max #6 overall (1193), +12 points over MAI-Image-2 - UNI-1.1 #7 overall (1190), +13 points over Reve-v1.5 Multi-Image Edit Arena - UNI-1.1-Max #7 overall (1315), +21 points over Seedream 4.5 - UNI-1.1 #8 overall, (1298) Single-Image Edit Arena - UNI-1.1-Max #7 overall (1337) - UNI-1.1 #11 overall, (1310) on par with Grok-Imagine-Image (20260207) Congratulations to @LumaLabsAI on this solid performance!
OpenRouter@OpenRouter精选AI 评分6565推荐理由：OpenRouter 拆解了 GPT 5.5 的实际成本，49-92% 的涨价被输出 token 减少部分抵消，用 API 的人必须算的一笔账。
IBM近日发布了采用Apache 2.0许可证的Granite 4.1大语言模型系列。随后，Unsloth提供了该系列3B模型的21个GGUF量化变体。作者利用这些大小从1.2GB到6.34GB不等的量化模型，尝试生成“鹈鹕骑自行车”的SVG图像以测试其能力。实验结果显示，所有模型生成的图像质量均较差，且未观察到模型大小与输出质量之间存在明显关联。基于此次不理想的尝试，作者表示未来将选用更擅长图像生成的模型重新进行此类实验。
Elon Musk@elonmuskAI 评分4141Grok 4.3 just became the smartest AI in the world at law and money It took #1 on TWO brutal private tests no other model could win on “Vals AI” benchmarks #1 CaseLaw (v2) - 79.31% accuracy Private Q&A benchmark over real Canadian court cases. Tests deep legal reasoning, precedent understanding, and precise answers from complex judgments. (outranking GPT-5.1 at 73.42%) #1 CorpFin (v2) - 68.53% accuracy Private benchmark on long-context credit agreements. Evaluates how well models truly understand dense, multi-page financial contracts, terms, risks, and clauses These are not just basic tests - they’re real-world, high-stakes legal + financial reasoning challenges Grok 4.3 leads in accuracy on both, proving it’s not just fast or cheap… it’s the smartest at the hardest real world tasks xAI is building the reasoning engine the world needs
Ethan Mollick@emollickAI 评分5757美国政府机构评估称中国在人工智能竞赛中落后八个月，但独立数据并未证实这一差距。当前美国实验室持续追求更智能的模型，而中国玩家如深度求索（Deepseek）等提供的价格优势可能成为更关键的竞争筹码。这场竞赛的衡量标准正从单纯的技术指标扩展到包括成本效益在内的综合维度。
一项新基准测试让领先的语言模型处理100个日常伦理场景，涵盖从销售数据滥用到肿瘤学违规操作等领域。测试结果显示，不同前沿模型对相同伦理提示给出了差异显著的回应。这引出了一个核心问题：究竟由谁来决定AI被允许做什么，以及它应遵循谁的伦理准则？该基准旨在揭示和量化主流AI系统在道德判断上的不一致性。
ARC Prize Foundation 对 OpenAI 的 GPT-5.5 和 Anthropic 的 Opus 4.7 在 ARC-AGI-3 基准测试中的 160 次任务运行进行了分析。研究发现，三种系统性错误模式导致这两个模型在人类能轻松解决的任务上得分均低于 1%。这些错误揭示了当前顶尖大语言模型在抽象推理能力上仍存在根本性缺陷。
Elon Musk@elonmuskAI 评分3939Grok Voice brutally dominates the top of the τ-voice Bench Grok scores 67.3%, while Gemini sits at 43.8% and GPT Realtime at 35.3% This is a massive lead over the competitors and it's not even close The best real-time reasoning voice agent out there
TestingCatalog News 🗞@testingcatalogAI 评分6666🚨 Google updated Gemini 3 Flash in arena It still has the same name "Gemini 3 Flash". However, output quality is two tiers above it. It could be 3.1, 3.2 or 3.5 Flash. Not sure what they'll call it. It's performance is closer to current 3.1 Pro than the current 3 Flash. Huge upgrade.
François Chollet@fcholletAI 评分5656最新一批模型在ARC-AGI-3上的得分目前仍低于1%。 到今年年底，得分会达到多少呢？
GPT-5.5 & Opus 4.7 on ARC-AGI-3 - GPT-5.5: 0.43% - Opus 4.7: 0.18% We found 3 failure modes: - True local effect, false world model - Wrong level of abstraction from training data - Solved the level, didn’t reinforce the reward See our full analysis 🧵
François Chollet@fchollet精选AI 评分7070GPT-5.5 在 ARC AGI 3 上得分 0.43%！ - GPT-5.5：0.43% - Opus 4.7：0.18% - GPT-5.4：0.20% - Claude 4.6：0.45% - Gemini 3.1：0.4% GPT 5.5 报告的失败原因包括： - 真实的局部效果，错误的世界模型 - 来自训练数据的抽象层级错误 - 解决了关卡，但没有强化奖励 我认为完整的分析将帮助 OpenAI 全面理解模型在某些模态中失败的地方
推荐理由：Chollet 用 ARC AGI 3 冷冰冰的数字撕开了 RL 的局限，GPT-5.5 0.43% 的得分说明在未知领域模型会做完全不相干的事，比任何安全论文都来得更直击要害。
PixVerse@PixVerse_AI 评分4949SeeDance 2.0 vs HappyHorse 1.0 Very quick text-to_video comparison. Which do you prefer? Universal Prompt: "Aesthetic: lofi, cozy, warm tones, soft film grain, gentle bloom, analog textures. Camera default: 35 to 50mm equivalent, shallow depth of field, slow subtle moves, cut length 0.6 to 1.2 seconds. Style: Cozy cel-shaded anime." Thanks to @PixVerse_ for HappyHorse generations which are currently free to members. lofi audio tracks were pretty poor for both so I used @Suno for audio.
英国 AI 安全研究所评估发现，OpenAI 的 GPT-5.5 成为第二个能够自主完成完整网络攻击模拟的 AI 模型。其表现与 Anthropic 的 Claude Mythos 几乎持平，后者目前仍仅限小范围使用。GPT-5.5 现已通过 ChatGPT 和 API 向公众广泛提供。
TestingCatalog News 🗞@testingcatalogAI 评分5555Grok 4.3 在 Artificial Analysis Index 中升至第 7 位，超越了 Meta 的 Muse Spark。
This release shows increased cost efficiency to run the Artificial Analysis Intelligence Index, with Grok 4.3 sitting comfortably on the Pareto frontier for intelligence versus cost Driven by 37.5% lower input token prices and 58.3% lower output token prices, it costs $395 to run the Intelligence Index evaluations, an overall ~20% decrease from Grok 4.20 0309 v2
Artificial Analysis@ArtificialAnlysAI 评分5757
Mark Chen@markchen90AI 评分3939OpenAI’s GPT-5.5 is the second model to complete one of our multi-step cyber-attack simulations end-to-end 🧵
Ethan Mollick@emollickAI 评分6161xAI has launched Grok 4.3, achieving 53 on the Artificial Analysis Intelligence Index with improved agentic performance, ~40% lower input price, and ~60% lower output price than Grok 4.20 The release of Grok 4.3 places @xAI just above Muse Spark and Claude Sonnet 4.6 on the Intelligence Index, and a 4 points ahead of the latest version of Grok 4.20. Grok 4.3 improves its Artificial Analysis Intelligence Index score while reducing cost to run the benchmark suite. Key Takeaways: ➤ Grok 4.3 improves on cost-per-intelligence relative to Grok 4.20 0309 v2: it scores higher on the Intelligence Index while costing less to run the full benchmark suite. Grok 4.3 costs $395 to run the Artificial Analysis Intelligence Index, around 20% lower than Grok 4.20 0309 v2, despite using more output tokens. This makes it one of the lower-cost models at its intelligence level ➤ Large increase in real world agentic task performance: The largest single benchmark improvement is on GDPval-AA, where Grok 4.3 scores an ELO of 1500, up 321 points from Grok 4.20 0309 v2’s score of 1179 Grok 4.3, surpassing Gemini 3.1 Pro Preview, Muse Spark, Gpt-5.4 mini (xhigh), and Kimi K2.5. Grok 4.3 narrows the gap to the leading model on GDPval-AA, but still trails GPT-5.5 (xhigh) by 276 Elo points, with an expected win rate of ~17% against GPT-5.5 (xhigh) under the standard Elo formula ➤ Grok 4.3’s performs strongly on instruction following and agentic customer support tasks. It gains 5 points on 𝜏²-Bench Telecom to reach 98%, in line with GLM-5.1. Grok 4.3 maintains an 81% IFBench score from Grok 4.20 0309 v2 ➤ Gains 8 points on AA-Omniscience Accuracy, but at the cost of lower AA-Omniscience Non-Hallucination Rate of 8 points, so Grok 4.20 0309 v2 still leads AA-Omniscience Non-Hallucination Rate, followed by MiMo-V2.5-Pro, in line with Grok 4.3 Congratulations to @xAI and @elonmusk on the impressive release!
OpenRouter@OpenRouter精选AI 评分6868推荐理由：Grok-4.3 降价但性能反升，agentic 跑分直接到 1500，如果之前觉得 Grok 贵而没试过，这次可以上车了。
Rohan Paul@rohanpaul_aiAI 评分5858OpenAI’s GPT-5.5 is the second model to complete one of our multi-step cyber-attack simulations end-to-end 🧵
Artificial Analysis@ArtificialAnlysAI 评分6565Artificial Analysis@ArtificialAnlysAI 评分4646
Sam Altman@samaAI 评分4343GPT-5.5 is on par with Claude Mythos - GPT-5.5 average pass rate of 71.4% (±8.0%) - Mythos Preview 68.6% (±8.7%) - GPT-5.5 solved a task that takes a human expert ~12 hours in under 11 minutes at a cost of $1.73
