In the WDCD v3.1 compliance test, Grok 4 ranked first with 91.04 points while Doubao Pro placed last with 58.00 points, a 33.04-point gap between the top and bottom.
Qwen3 Max's main leaderboard score in the Smoke evaluation fell from 96.04 to 85.82 in a single day, a decline of 10.2 points. The material constraint dimension accounted for most of the drop, falling 21.5 points, while code execution remained nearly unchanged.
DeepSeek V4 Pro's material constraint score fell from 89.50 to 70.00 in today's Smoke evaluation, dragging the main ranking down from 94.07 to 86.50. The drop is likely attributable to sampling variance rather than systematic model degradation.
During August 6–9, 2026, Claude Sonnet 4.6 posted the largest single-model decline in the Smoke rapid test, falling 22.6 points, while GPT-o3 rose 6.7 points. GLM-4.6 recorded extreme volatility of 59.9.
On 2026-08-09, the YZ Index Smoke quick test covered 9 models, with GPT-o3 ranking first at 95.91 points. The test focuses on code execution and material constraint, with several top models showing notable declines.
Doubao Pro scored "-" on all five dimensions—execution, grounding, judgment, integrity, and communication—in today's Smoke evaluation, as well as on the main leaderboard. The model completed no tasks due to API failure/timeout and is excluded from this period's ranking.
In today's Smoke evaluation, Claude Sonnet 4.6's code execution score dropped from 94.50 to 75.00, a decline of 19.5 points, while its material constraint score surged from 43.30 to 97.80, lifting the overall leaderboard score from 71.46 to 85.26.
On 2026-08-08, the YZ Index Smoke quick test covered 9 models, with Claude Opus 4.7 and GPT-o3 tying at 97.66 points for the top spot. Smoke is a daily 10-question quick test for observing short-term signals and does not carry the weight of the Full weekly leaderboard.
SpecForge v0.3.0: a Unified Disaggregated and Colocated Speculative Decoding Stack, and New Open SpecBundle Draft ModelsThe SpecForge TeamAugust 4, 2026When we first released SpecForge, a training job
Full-Stack Performance Optimization of AR+DiT in SGL-DiffusionAscend TeamAugust 05, 2026TL;DR Replaces the HF backend with SRT to accelerate AR modeling and resolve parallelism conflicts, with dedicat
HPC-Ops × SGLang: High-Performance Attention, Router GEMM, and MoE Kernels from Tencent HunyuanTencent Hunyuan AI Infra and the SGLang TeamAugust 7, 2026HPC-Ops is an open-source operator library for
Claude Opus 4.7's code execution score fell from 100.00 to 69.50 in today's Smoke evaluation, while the main leaderboard dropped only 6.4 points. The opposing movement of dimensions points to random question sampling rather than model degradation.
Grok 4's material constraint score fell from 82.60 to 65.00 in today's Smoke evaluation, while its overall main leaderboard score only declined from 82.99 to 81.23.
The 2026-08-07 YZ Index Smoke quick test covered 9 models, with Gemini 2.5 Pro ranking first at 87.21 points. As a small-sample daily signal, the results are better suited for short-term monitoring than long-term conclusions.
GLM-4.6 scored 71.90 on material constraint in today's Smoke evaluation but missed both the code execution and integrity dimensions due to API failures or timeouts, excluding it from main leaderboard ranking.
Doubao Pro's Smoke evaluation this cycle showed complete data loss across all five dimensions — execution, grounding, judgment, integrity, and communication — caused by an API failure or timeout. The model will not be included in the main leaderboard ranking this cycle.
The 2026-08-06 YZ Index Smoke quick test covered 9 models, with Claude Sonnet 4.6 and DeepSeek V4 Pro tying for first place at 92.17 points. Doubao Pro and GLM-4.6 were not ranked due to incomplete data from API failures or timeouts.
This round of WDCD v3.1 testing shows GPT-o3 rising 9.5 points and Gemini 2.5 Pro rising 7.6 points, while GLM-4.6 plunges 14.9 points, Claude Sonnet 4.6 drops 10.8 points, and Claude Opus 4.7 drops 5.9 points.
The WDCD v3.1 compliance test shows safety compliance as the hardest scenario, with gpt-5.5 and qwen3-max scoring only 1.8/4, while all 11 models in the engineering standards scenario scored at least 3.2/4. The results reveal a capability ceiling in engineering standards and a persistent gap in safety compliance.
Under a sampling scope targeting only 8 v2 anchor questions, 11 models posted an average R1 confirmation rate of 0.91, an average R2 resistance rate that fell to 0.68, and an average R3 integrity rate of just 54.5%. This trajectory reveals systematic attrition of constraints under sustained pressure.
