Claude Sonnet 4.6 rose 15 points in the latest WDCD v3.1 test compared to Run #227, while GLM-4.6 dropped 15.3 points, creating the most prominent compliance fluctuation signal among the 11 evaluated models.
WDCD v3.1 pilot data reveals business rules as the lowest survival dimension across 11 models, with champion deepseek-v4-pro scoring 4/4 and last-place qwen3-max only 1.55/4, showing significantly higher differentiation than the other four constraint categories.
In the WDCD v3.1 pilot, eight v2 anchor questions underwent three rounds of testing, showing all 11 models achieved 100% confirmation in the R1 constraint injection phase and 100% resistance in the R2 interference phase. However, entering the R3 pressure phase, the average integrity rate fell to 36.4% (out of a maximum score of 2), with four complete collapses (score 0) recorded.
GPT-o3 tops the WDCD v3.1 Commitment Ranking with 94.00 points, while Qwen3 Max ranks last with 62.60 points, showing a 31.4-point gap between the leading and trailing models among 11 evaluated models.
In today's Smoke evaluation, Gemini 3.1 Pro's main score dropped by 8.5 points to 72.50, with code execution halved from 75 to 50. The fluctuation is attributed to small sample size and question sampling variation.
DeepSeek V4 Pro’s code execution score in today’s Smoke Benchmark dropped from 98.70 to 75.00, a decline of 23.7 points, pulling the main leaderboard from 91.46 to 86.25. The fluctuation likely stems from random question sampling rather than model degradation.
The 2026-07-15 YZ Index Smoke quick test covered 10 models, with Claude Opus 4.7, Gemini 2.5 Pro, and GPT-5.5 tied at 100 points for the top spot. Smoke is a daily 10-question quick test, suitable for observing short-term signals, and is not equivalent to the Full weekly ranking conclusions.
On July 13, 2026, Colorado voted on a bill banning the use of artificial intelligence to generate child pornography, with 17 Republican lawmakers and some Democrats voting against. The article covers the bill’s mechanisms, industry impacts, and precedents.
On July 13, an employee of the U.S. Government Efficiency Department submitted a script named agent.py to a public GitHub repository, leading to the exposure of a valid xAI Grok API key that could access at least 52 xAI large language models. The key remained active even after the script was taken down.
Meta launched the Muse Image feature this week, which automatically enables and references public Instagram accounts, leading to privacy concerns and its eventual shutdown within three days.
xAI has recently launched the Grok 4.5 model, which excels in coding and agent tasks and has been integrated into development tools like Cursor, with availability extended to European users. This release coincides with a wave of agent-focused AI launches, signaling a shift from answering questions to completing workflows.
Apple has filed a lawsuit in a California federal court accusing OpenAI and a former Apple AI engineer of colluding to illegally obtain technical data through an unpatched vulnerability in Apple's internal systems. The case highlights the tension between AI talent mobility and protection of proprietary hardware technology.
OpenAI officially released the GPT-5.6 model along with the GPT-Live voice feature and the ChatGPT Work agent tool. Sam Altman described GPT-5.6 as a baseline model that needs to be surpassed, highlighting the shift from dialogue to agent tools.
On July 14, 2026, the YZ Index Smoke Quick Test covered 11 models, with DeepSeek V4 Pro ranking first with 91.46 points. The test focuses on code execution and material constraint dimensions, serving as a short-term signal monitor rather than a definitive long-term evaluation.
Intel announced an additional €5 billion investment in its Leixlip campus in Ireland to upgrade facilities and expand production capacity for Intel Xeon 6 and next-generation Xeon server processors based on the Intel 3 process node.
Meta has halted its Instagram Muse Image feature after backlash over default use of public photos for AI generation, shifting consent burdens to users and drawing criticism from creators, unions, and regulators.
A contract between OpenAI and the Pentagon triggers a severe backlash from employees and users, with ChatGPT uninstall rates spiking 295% and many users switching to Claude. Anthropic, which refused the Pentagon's "any lawful use" clause, is praised for its ethical stance.
Meta launched the Muse Image image generation model in early July 2026, then quickly removed the feature allowing users to @mention public Instagram accounts as visual reference. The default-on setting sparked opposition from SAG-AFTRA.
This week's 361 translation tasks were completed by 3 models. A sample of 3 tasks underwent multi-model blind evaluation, with the best overall being gpt-o3 (average score 9/10).
Apple has filed a lawsuit against OpenAI, alleging that the AI company systematically obtained trade secrets related to Apple's unreleased technology through former employees to develop consumer hardware products. The lawsuit also highlights OpenAI's pivot to hardware, straining their previous partnership.
