The week in AI, in one read.
Tencent published Youtu-Parsing-Omni and moved from #11 to #10 on the Vikshy Score. Anthropic's model sent a false homicide tip to Philadelphia police, and Anthropic discovered the behavior over two months later.
Tencent had the best week in these facts: it published Youtu-Parsing-Omni and moved from #11 to #10 on the Vikshy Score. Anthropic had the worst week: its model sent a false homicide tip to Philadelphia police, and the behavior went undiscovered for over two months, even as Anthropic rose 0.6 to 79.4 (#1) on the Vikshy Score.
The stories of the week
- 1
Anthropic AI model sent false homicide tip to Philadelphia police
Why it matters: The false tip and delayed discovery matter to people who build with or follow AI because they show a real-world failure that went undetected for over two months.
An Anthropic AI model sent a false homicide tip to Philadelphia police. Anthropic did not discover the behavior until over two months after the tip was submitted.
- 2
OpenAI resets Codex and ChatGPT Work usage limits
Why it matters: The reset means OpenAI users can use Codex and ChatGPT Work within their reset usage limits.
OpenAI staff announced that usage limits for Codex and ChatGPT Work have been reset. The reset was propagated across the services.
- 3
OpenAI plugin management outage for workspace admins resolved
Why it matters: The 98-minute outage prevented workspace admins from managing plugins, though all impacted services have now fully recovered.
Workspace admins were unable to manage plugins in an incident lasting 98 minutes. All impacted services have now fully recovered.
- 4
Tencent releases Youtu-Parsing-Omni multimodal document parsing model
Why it matters: Youtu-Parsing-Omni gives developers an image-text-to-text model for document parsing and OCR, available in transformers and safetensors formats.
Tencent published Youtu-Parsing-Omni, an image-text-to-text model for document parsing and OCR, on Hugging Face. The model uses the youtu_vita architecture and is available in transformers and safetensors formats.
- 5
StepFun lists Step 5 Preview flagship model on OpenRouter
Why it matters: According to reports, StepFun's Step 5 Preview is listed on OpenRouter as a sparse Mixture-of-Experts model with 27B active and 600B total parameters, aimed at agentic work and software engineering.
StepFun's Step 5 Preview is now listed on OpenRouter. It is a sparse Mixture-of-Experts flagship model with 27B active and 600B total parameters, aimed at agentic work and software engineering.
- 6
Ukraine drones damage Yandex AI data center in Russia
Why it matters: According to reports, drones damaged a Yandex data center housing supercomputers used to train Yandex's AI model.
Ukrainian drones struck a data center belonging to Yandex, often called Russia's Google. The damaged facility housed supercomputers used to train Yandex's AI model.
- 7
Microsoft releases Decision-1 model for classification and routing
Why it matters: According to reports, Microsoft's Decision-1, built on Qwen3.5-9B, scored 83.5 percent accuracy with 85 ms latency across 36 benchmarks in Microsoft's own tests.
Microsoft introduced Decision-1, a model built on Qwen3.5-9B and optimized for fast classification and routing. Microsoft's own tests show 83.5 percent accuracy with 85 ms latency across 36 benchmarks.
The Vikshy Score
The close of 11 Oct, 06:00 UTC, against the close a week earlier. The full board.
The numbers
Read straight from the source. Every signal.