On June 30, 2026, Anthropic announced the resumption of Claude Fable 5 global service from July 1, ending a nearly three-week suspension due to US export controls. It released a new safety classifier intercepting over 99% of jailbreak attempts and established an AI jailbreak evaluation framework with Amazon, Google, Microsoft, and the US government.
The "i have adhd" prompt technique, shared by developer jacky, significantly improves Claude's response quality by forcing the model to present answers directly, and has sparked widespread discussion in the developer community.
A post directly quoting "China is leading the AI race" with accompanying laughter sparks widespread discussion, based on the actual performance of models like Kimi K3 and DeepSeek, challenging assumptions about hardware dependency. The debate reflects shifting views on whether low-cost paths can produce competitive AI outcomes, moving the focus from absolute scale to efficiency.
The world faces severe computing power constraints, with demand far outpacing supply expansion despite limited AI adoption. Physical infrastructure bottlenecks and capital expenditure disparities are reshaping industry competition, leading to resource concentration among top players.
Grok 4.5 scored 91.3% on the new VulcanBench coding benchmark, solving 21 real-world software tasks across five languages, outperforming Claude Fable 5 and GPT-5.6 Sol while maintaining cost efficiency. Elon Musk's "Yes" response to a suggestion to try Grok 4.5 further fueled discussions.
Moonshot AI has temporarily suspended new subscriptions for Kimi K3 due to GPU resource pressure and demand reaching capacity limits, prioritizing existing subscribers while planning to split membership into two plans.
In today's Smoke evaluation, Qwen3 Max's main score dropped from 82.23 to 67.31, a decrease of 14.9 points, with the code execution dimension falling from 96.90 to 65.60.
In today's Smoke evaluation, Gemini 2.5 Pro's code execution score dropped from 74.60 to 50.00 points (a decrease of 24.6 points), while its overall ranking fell from 76.49 to 69.98 points.
On July 20, 2026, the YZ Index Smoke Quick Test covered 11 models, with Claude Opus 4.7 scoring 100 points to top the daily rankings. This Smoke test uses 10 questions per day, suitable for observing short-term signals, and is not equivalent to the Full weekly ranking conclusions.
A recent Science Magazine report reveals that an AI system has constructed an entirely new language, challenging traditional views on human-exclusive creativity and triggering discussions across industry and research fields.
On July 18, 2026, the HumansFirst organization initiated 142 protests across 42 states, marking the first nationwide coordinated action against AI infrastructure expansion. The protests target data center projects by Meta, Alphabet, Amazon, Microsoft, and xAI, highlighting concerns over water consumption, electricity costs, noise pollution, and farmland loss.
On July 16, 2026, 29 countries signed an agreement in Shanghai to formally establish the World Artificial Intelligence Cooperation Organization (WAICO), headquartered in Shanghai, aiming to promote international AI cooperation and global governance.
In July 2026, Thinking Machines Lab released the Inkling model with 975 billion total parameters and 41 billion activated parameters, supporting text, image, and audio inputs with a 1 million token context. The model leads among U.S. open-weight models in agent tasks but lags behind top Chinese models overall.
On July 16, 2026, Moonshot AI (Beijing) released the Kimi K3 model with 2.8 trillion parameters and a 1 million token context window, making it the world's largest open-source weights model.
In July 2026, Inio Asano paused the serialization of "Mujina Into the Deep" due to publisher restrictions on generative AI for professional artists, exposing tensions between copyright concerns and creative efficiency.
In today's Smoke evaluation, Gemini 3.1 Pro's material constraint score fell from 90.90 to 64.30, a drop of 26.6 points, while its main ranking overall still rose by 5.4 points.
GPT-o3's main score in today's Smoke evaluation fell from 80.61 to 66.86, with code execution dropping from 70.30 to 48.50, a single-day decline of 21.8 points.
The seven-day Smoke evaluation from July 13 to July 19, 2026 shows Claude Opus 4.7 leading with an average score of 86.9, while GPT-o3 declined sharply by 30.5 points. The report also analyzes rising models such as Qwen3 Max, declining models like Doubao Pro, along with volatility and integrity ratings.
On 2026-07-19, the YZ Index Smoke Quick Test covered 10 models, with Claude Opus 4.7 ranking first at 95.19. Smoke is a daily 10-question quick test suitable for monitoring short-term signals and is not equivalent to Full weekly rankings.
xAI hired two researchers from NVIDIA in the summer of 2025 to work on world model development, planning to first apply it to generating interactive 3D game environments and release an AI-generated game by the end of next year.
