The day in AI, in five minutes.
OpenAI says it disrupted a campaign to extract protected model reasoning, while Google announced SynthID Bio for watermarking AI-generated proteins. Google also reports up to 1.69x end-to-end inference speedup for 1440p video generation on TPUs; Gemini 4 Argon enters Arena's text leaderboard at #1, and several lab rankings and model prices change.
What mattered
- 1
OpenAI disrupts coordinated model-distillation campaign
Why it matters: The reported campaign and OpenAI's response matter to builders who need to protect model reasoning from extraction.
OpenAI says it disrupted a campaign to extract protected model reasoning and is strengthening defenses against adversarial distillation.
- 2
Google introduces SynthID Bio for watermarking AI-generated proteins
Why it matters: SynthID Bio gives people working with AI-generated proteins a proof of concept for watermarking them while preserving biological function.
Google announced SynthID Bio, a proof of concept for watermarking AI-generated proteins while preserving biological function.
- 3
IQuest Research open-sources IQuest-Q1 coding model
Why it matters: According to reports, IQuest Research released open weights for IQuest-Q1, a 320B sparse MoE model with 15B active parameters, a 512K context window and a score of 84.5 on CyberGym, for command-line coding agents.
IQuest Research released open weights for IQuest-Q1, a 320B sparse MoE model with 15B active parameters built for command-line coding agents. It has a 512K context window and scores 84.5 on CyberGym.
- 4
DeepSeek open-sources six software modules for Huawei Ascend AI chips
Why it matters: DeepSeek's six open-source modules give developers tools tailored to Huawei's Ascend AI chips as the company aims to build an independent software ecosystem for GPUs.
DeepSeek released six open-source software modules tailored for Huawei's Ascend AI chips, mirroring its earlier tools for Nvidia GPUs. The company says the aim is to build an independent software ecosystem for GPUs.
- 5
Google speeds up video diffusion attention on TPUs
Why it matters: Sparse VideoGen and Splash Attention kernel optimizations show a way to speed up 1440p video generation inference on TPUs, with up to 1.69x end-to-end speedup.
Google developers implemented Sparse VideoGen to route attention heads to structured sparse masks for video diffusion. Combined with Splash Attention kernel optimizations, this achieved up to 1.69x end-to-end inference speedup for 1440p video generation on TPUs.
The Vikshy Score
The close of 1 Oct, 06:00 UTC, against the previous close. The full board.
The numbers
Read straight from the source. Every signal.