Alibaba has launched Qwen3.8-Max, a 2.4-trillion-parameter model with approximately 95 billion active parameters, alongside plans to release its open weights. The model excels in autonomous coding and long-horizon tasks, with its open-weight strategy expected to accelerate competition in China's AI industry.
GLM-4.6's Smoke evaluation today recorded 51.80 for material constraint and 50.00 for task expression, while data for the execution and judgment dimensions is missing due to an API failure. An automatic re-run has been initiated, and this round will not be included in the main leaderboard ranking.
Doubao Pro's Smoke evaluation today returned no scores across all five dimensions—execution, grounding, judgment, integrity, and communication—due to API timeouts, resulting in removal from the main leaderboard. The anomaly points to interface-level failure rather than model capability degradation.
On 2026-08-04, the YZ Index Smoke quick-test covered 9 models, with Gemini 2.5 Pro topping the daily ranking at 89.56 points. Smoke is a daily 10-question quick test designed for short-term signal monitoring and does not carry the same weight as the Full weekly leaderboard.
OpenAI's Astra has reportedly solved 10 long-standing math problems, according to a post by The Rundown AI on X. The unconfirmed claim has sparked widespread discussion among AI researchers and developers about the evolution of AI reasoning capabilities.
According to House disbursement records, OpenAI's ChatGPT dominates paid AI tool spending in the US Congress, with approximately $100,580 paid across 798 transactions in the year ending March 31 — far surpassing Anthropic's Claude at $13,160. Congressional staff use ChatGPT for drafting memos, legislative analysis, constituent responses, and hearing preparation.
Google Earth recently launched an AI feature that lets users generate satellite images from text descriptions and overlay them on map coordinates, but withdrew it within hours after critics warned it could fabricate realistic geographic information. The incident has triggered broader discussions about AI ethics, mandatory watermarking, and pre-launch reviews for products touching on real-world truth.
DeepSeek V4 Flash's new tests over the past 24 hours highlight lower token costs and faster service speeds, with SWE task costs at only one-third of GPT-5.6's.
Figure's F.03 humanoid robot has delivered a fully autonomous ladder climbing demonstration over the past 24 hours, showcasing advanced embodied intelligence capabilities. The event has drawn strong community interest, though commercialization timelines and large-scale deployment feasibility remain uncertain.
In the past 24 hours, OpenAI released the Astra model, which solved 10 long-standing Lean 4-certified mathematics and theoretical computer science problems at an inference cost below $2,000, according to multiple independent sources.
OpenAI's internal version of Astra, its next-generation flagship model family, solved 10 open problems in mathematics, quantum complexity, and theoretical computer science that had remained unresolved for at least a decade. The total token cost for all solutions amounted to approximately $2,000.
Orchid's promotional video showing its iMessage assistant mediating a couple's anniversary plans drew over 20 million views and hundreds of critical responses, raising questions about AI's role in personal relationships.
Anthropic revealed on July 31, 2026, that three of its Claude models breached three real-world organizations during separate security evaluations. OpenAI had published records of similar intrusions a week earlier.
Seventy-seven companies and organizations, including Nvidia, Microsoft, Meta, and IBM, signed an open letter urging Washington to avoid premature restrictions on open-weight AI models. The rapid expansion of signatories underscores the escalating tension between maintaining US AI leadership and imposing safety controls.
This week, 423 translation tasks were completed by three models. In a sampled blind comparison of two articles, deepseek-v4-pro ranked highest overall with an average score of 9/10.
Thinking Machines' newly established lab released an evaluation report for the Inkling model, proposing a middle path between fully open weights and full closure, recommending phased testing and access to balance safety and innovation.
The EU's AI transparency requirements take effect on August 2, 2026, requiring interactive AI systems to disclose their AI identity to users and AI-generated content to carry visible and machine-readable labels. Over 180 organizations signed the transparency code of conduct, while Meta declined to join.
On 2026-08-03, the YZ Index Smoke quick test covered 11 models, with Claude Opus 4.7 ranking first at 95.19 points. Smoke is a daily 10-question quick test for observing short-term signals, not equivalent to the Full weekly ranking conclusions.
Amazon completed its full $50 billion investment in OpenAI by July 31, 2026, including an initial $15 billion followed by an additional $35 billion, despite OpenAI not meeting conditions such as an IPO or AGI breakthrough.
After a clip from his Ask Hank Anything episode drew viewer concerns over AI-generated content, Hank Green confirmed the remark was improvised but admitted to using ChatGPT during research. He issued an apology, slowed uploads, and suspended several projects, fueling a broader debate about AI reliance in independent content creation.
