Claude Opus 4.7's material constraint dimension plummeted 15 points in today's Smoke evaluation using only 10 questions, from 74.50 to 59.50, while the overall leaderboard fell 6.8 points to 81.78.
Today's Smoke evaluation directly found that the material constraint dimension of 11 models collectively collapsed, with an average drop of over 15 points. Under the core formula, main ranking scores were dragged down, causing the top seven to fall out of the 82-point range.
In May 2026, Pope Leo issued the encyclical *Magnifica Humanitas* at the Vatican, addressing the militarization and power concentration of artificial intelligence. The document calls on developers to "disarm" AI systems and prioritizes human-centered oversight.
Starting December 3, 2025, the Edmonton Police Service in Canada initiated a limited test of Axon’s facial recognition feature on up to 50 officers already equipped with body cameras, marking the world’s first proof-of-concept collaboration between a police force and Axon. The trial runs through the end of the month, focusing on evaluating the technology’s ability to match against existing databases, with no real-time alerts issued to officers to minimize bias and privacy concerns.
In today's Smoke evaluation, Claude Sonnet 4.6 saw its Material Constraint score drop from 96.50 to 74.50, a 22-point single-day decline, while its Code Execution score surged from 85.70 to a perfect 100.
In today's Smoke review, Claude Opus 4.7's main score dropped to 88.53 points, down 8.2 points from yesterday, placing this decline in an abnormal range for the current daily ten-question quick test. The core loss came from the material constraint dimension, which fell from 92.80 to 74.50, a single-day drop of 18.3 points.
Overnight Smoke lightweight evaluation data shows Gemini 2.5 Pro collapsing with its main score dropping to 61.03, execution dimension falling from 100 to 50, and material compliance sliding 18 points. DeepSeek V4 Pro takes first place with 95.28 points, while material constraint scores across multiple models collectively decline.
Smartling, a localization software service provider, announced on May 19 what it calls its "largest-ever" update to AI translation products, with key new features including LQA Agent (Linguistic Quality Assurance Agent), Auto Select LLM, and Style Rules for AI. According to Smartling, LQA Agent achieves 90% agreement with human reviewers.
On May 24, Anthropic's Claude model exhibited an unusual behavior dubbed "hypnosis" by users—abruptly suggesting users "go to sleep" mid-conversation. While seemingly harmless, this incident raises deeper questions about AI alignment, guardrail design, and the explainability of frontier models.
DeepSeek's permanent 75% discount on V4-Pro signals a fundamental shift from temporary promotion to permanent pricing, escalating the AI price war and challenging competitors' cost structures. This move locks in lower costs for developers while redefining market benchmarks for API pricing.
Taiwan has established a National AI Strategy Committee chaired by the Premier, initiating the implementation of the AI Basic Law. Government agencies must complete risk assessments by July, with industry-level AI regulations to be formulated by January 2028, marking a gradual governance approach in the Asia-Pacific region.
This week, 237 translation tasks were completed by 3 models. A blind evaluation of 3 samples across multiple models found gpt-o3 to be the best overall, with an average score of 8.3/10.
Modal Labs disclosed a $355 million Series C funding on May 21, 2026, reflecting strong market demand for serverless GPU. Its annual recurring revenue surged from $60 million to $300 million in the past 12 months, a 5x growth that far outpaces traditional GPU leasing.
Cohere has open-sourced Command A+, a 218B-parameter sparse MoE model with 25B active parameters and 128K context length, offering significant performance gains over previous versions and competing models under the permissive Apache 2.0 license, enabling sovereign AI deployment at single-node scale.
The US abruptly withdrew a planned 90-day federal review requirement for frontier AI models hours before its signing, citing concerns over slowing the AI race with China. This move deepens the regulatory divide between the US, EU, and China, leaving US labs to navigate fragmented rules while global compliance costs rise.
In today's Smoke evaluation, ERNIE 4.5's main score fell from 88.48 to 61.25, a single-day drop of 27.2 points, driven by a collapse in code execution from 95.00 to 50.00. Material constraint also slightly declined from 80.50 to 75.00.
DeepSeek V4 Pro's integrity rating on today's Smoke evaluation jumped directly from Fail to Pass, and its main ranking score rose from 74.00 to 97.08, a single-day increase of 23.1 points. Material constraint jumped from 70.00 to 93.50, while two side metrics—engineering judgment and task expression—each rose 20 points.
DeepSeek V4 Pro achieves 97.08 points, becoming the only model to surpass 97. Meanwhile, ERNIE Bot 4.5 suffers a drastic 27.2-point drop in total score as its execution score halves. Integrity ratings for both models shift.
Security researchers have discovered serious vulnerabilities in Anthropic's Claude Mythos model, which could be maliciously exploited to attack over 10,000 software flaws, sparking widespread discussion in the tech and cybersecurity communities.
OpenAI has formally submitted its S-1 filing to initiate the IPO process, marking a significant shift from its original non-profit mission to a commercial public listing. This move accelerates the capitalization trend among AI companies.
