WDCD Run #202 (2026-06-28) measured multi-turn commitment integrity across 11 frontier models, recording an average instruction decay of -73.2% between Round 1 and Round 3. Gemini 3.1 Pro topped the leaderboard at 93.6 points.
In the latest WDCD cycle (Run #196), all eight evaluated models showed positive changes, with none declining. Claude Opus 4.7 recorded the largest single-model increase of 19.8 points, jumping to 89.29 points and entering the top three.
In the WDCD Compliance Test, the safety compliance scenario scored the lowest on average across all models, with the highest score being only 3.57/4 for deepseek-v4-pro, while claude-sonnet-4.6 scored only 2.57/4.
In WDCD tests, Grok 4 maintains a 1.83/2 honesty rate with zero crashes in R3, while both Claude Sonnet 4.6 and GPT-o3 suffer six complete R3 crashes each (17.1%). The data reveals systematic degradation across three pressure rounds, with overall honesty dropping to 1.63/2 in the final stage.
Gemini 3.1 Pro leads the WDCD compliance ranking with 93.57 points (R1=1.00, R2=0.97, R3=1.77/2), while ERNIE Bot 4.5 ranks 11th with 75.71 points (R1=0.89, R2=0.60, R3=1.54/2).
In the June 2026 Smoke review of the YZ Index, Claude Sonnet 4.6's main score fell from 96.45 to 70.52, code execution dropped from 100.00 to 50.00, while material constraint rose from 92.10 to 95.60.
In today's Smoke evaluation of the YZ Index, Claude Opus 4.7 saw its main score drop from 97.12 to 71.47, a decline of 25.7 points, driven entirely by the code execution dimension halving from 100.00 to 50.00.
In the YZ Index Smoke tests from June 23 to 28, 2026, ERNIE Bot 4.5 showed the largest decline, dropping 37.2 points from 98.74 to 61.52, with an average of 82.1, while multiple models experienced fluctuations exceeding 28 points. Only Doubao Pro maintained a slight upward trend, emerging as the sole stable performer.
In the Smoke lightweight benchmark on June 28, 2026, Doubao Pro topped the main leaderboard with 98.61 points (Execution 100, Material Constraint 96.9). Claude Opus 4.7 and Sonnet 4.6 saw their Execution scores drop from 100 to 50, causing significant ranking declines.
OpenAI and Broadcom jointly announced the launch of Jalapeño, the first custom ASIC chip optimized for large language model inference, aiming to reduce per-response costs by approximately 50% and lessen reliance on NVIDIA. Scheduled for deployment by end of 2026 and mass production by 2027-2028.
Anthropic sent letters to Reuters and the U.S. Congress on June 24-25, 2026, accusing Alibaba affiliates of using approximately 25,000 fake accounts to generate over 28.8 million Claude interactions from April 22 to June 5, attempting to distill its reasoning and programming capabilities.
Hasbro has added an AI voice replication clause to Peppa Pig renewal contracts, requiring child stars to agree to permanent use of their voices for commercial assets. The British Child Star Agency Association has publicly opposed this.
The artificial intelligence wave is profoundly reshaping the supply-demand landscape of the semiconductor industry. Following strong earnings from memory chip giant Micron Technology, related chip stocks have surged, reigniting market enthusiasm for AI data center construction.
The rise of AI agents is transforming sales automation, with tools like Claude Sales enabling one-click outreach and automated follow-ups. The trend marks a shift from model training to practical application, though concerns over data privacy and reliability remain.
Several employees of AI company Anthropic have publicly shared psychological distress after using the company's AI product Claude, questioning the value of their own work and experiencing mild depressive symptoms. This phenomenon has sparked heated debate in the tech industry about AI replacing the meaning of human work.
OpenAI recently announced delaying the public release of its GPT-5.6 model due to U.S. government national security concerns over frontier AI technology, igniting widespread debate on AI regulation and innovation freedom.
OpenAI has officially announced the launch of its first custom AI inference chip in partnership with semiconductor giant Broadcom, a breakthrough expected to reduce data center operational costs by up to 50% and support larger-scale AI model deployments.
In the YZ Index from June 27, 2026, Smoke lightweight evaluation, Claude Opus 4.7 ranked first on the main leaderboard with 97.12 points, achieving a perfect 100 points in code execution and 93.6 points in material constraint.
On June 25, 2026, OpenAI confirmed that GPT-5.6 would only be available to a small group of partners, each subject to individual U.S. government review, directly limiting early access to the model and fueling debate between safety and innovation advocates.
From June 24 to 25, 2026, the U.S. government demanded that Anthropic fully suspend global access to Fable 5 and Mythos models on grounds of export controls and national security, escalating tensions between safety oversight and AI competitiveness.
