AI News

2026-09-11 · FridayPublished daily at 08:00

AI Daily

Issue 143112026-09Friday
16major stories12sources7firsthand announcements4new modelsAbout a 5-minute read
Today's Highlights
  1. 1
  2. 2
  3. 3
In This Issue
  1. 01Model Releases/Updates4 items
  2. 02Product Releases/Updates6 items
  3. 03Industry News2 items
  4. 04Paper research1 item
  5. 05Tips and perspectives3 items

01Model Releases/Updates

DeepSeek: API ChangelogFirsthand

DeepSeek releases V4.1-Flash, with API prices lowered simultaneously

DeepSeek releases DeepSeek-V4.1-Flash, the smallest model in its new model architecture series, with native multimodal visual understanding. Its evaluations include GPQA Diamond 90.9, HLE 36.8, Codeforces Rating 3471, and Terminal-Bench 2.1 90.6.

Original text
Tencent WorkBuddy

WorkBuddy launched DeepSeek V4.1-Flash, with a two-week free trial.

WorkBuddy announced that DeepSeek V4.1-Flash is now available on its platform, with a two-week free trial. The cited DeepSeek announcement says V4.1-Flash is now available through the DeepSeek API and supports native multimodality.

Original text
SiliconFlow

DeepSeek-V4.1-Flash launches on SiliconFlow, with 552B MoE and support for a 1M context window

SiliconFlow announced that DeepSeek-V4.1-Flash went live on its platform on Day 0. The model is a 552B MoE with approximately 8B active parameters during prefill and approximately 16B during decode, native vision, a 1M context window, and KV cache usage of approximately one-quarter that of V4 Flash. It uses the MIT License.

Original text
Suno

Suno v6 launches, supporting music generation from images, video, and voice memos

Suno releases the v6 model, which can turn images, videos, and voice memos into music and make precise modifications to existing songs. It also offers the v6-wild version for exploring more possibilities, along with an official feature demo video under 2 minutes long.

Original text

02Product Releases/Updates

OpenAI: Website updatesFirsthand

OpenAI releases the public beta of the Agents API

OpenAI launched the public beta of the Agents API, making the harness and infrastructure that power Codex available to developers through a single API call, hosted in the cloud.

Original text
Cursor BlogFirsthand

Cursor introduces Projects: a coordinator agent manages thousands of subagents for large development tasks

Cursor releases Projects (beta), allowing users to handle large-scale work such as feature development, migrations, and ongoing maintenance through a coordinator agent. The coordinator does not write code itself; it dispatches thousands of sub-agents to execute tasks in parallel.

Original text
OpenAI: Website updatesFirsthand

OpenAI introduces the full-duplex voice model GPT-Live-1 in its API

OpenAI released the voice model GPT-Live-1 in its API. It can listen and speak simultaneously and supports delegating reasoning and tool calls to backend models such as GPT-6 Astra. The frontend voice layer is priced at $0.05 per minute.

Original text
Google AI

Google launches Pics, an image tool based on Nano Banana that supports precise editing and collaboration

Google releases Google Pics, an image generation tool built on Nano Banana, now available at pics.new. It supports localized object editing, modifying and translating text within images, collaborative creation by multiple people, and generating multiple options from a single prompt.

Original text
Hugging Face Community Blog

Hugging Face rebuilds Workflow1111 with Gradio Workflow, replicating the main features of AUTOMATIC1111

Hugging Face released Workflow1111, using gr.Workflow to rebuild most of AUTOMATIC1111's functionality with 73 nodes and 11 media pipelines. It covers text-to-image generation, high-resolution restoration, image-to-image, prompt matrix, VLM prompt reverse engineering, detection-generated inpaint masks, ControlNet-style preprocessors, background removal, PNG Info, and image-to-video.

Original text
OpenAI: Website updatesFirsthand

OpenAI launches Data agent in ChatGPT Work

OpenAI introduces a new Data agent in ChatGPT Work, allowing users to connect company data, analyze changes, and generate shareable interactive dashboards using natural language.

Original text

03Industry News

Shopify announces a full migration back from React Native to native Swift and Kotlin development

Original text
TechCrunch:AI

An Anthropic report accused Alibaba, Moonshot, and DeepSeek of launching distillation attacks against Claude.

Anthropic published a report accusing several Chinese AI companies of repeatedly conducting distillation attacks on Claude. It identified nearly 200 million related interactions across five campaigns.

Original text

04Paper research

Anthropic:ResearchFirsthand

Anthropic evaluates AI models’ tactical intelligence, targeting, and conventional weapons capabilities

Anthropic Frontier Red Team has released new evaluations measuring models’ capabilities in tactical intelligence localization (account linking and geolocating photos and text) and conventional weapons development (terminal guidance and delivery for drones, and navigation under GPS jamming).

Original text

05Tips and perspectives

The Decoder:AI News

Swarmchasers tracked suspected OpenAI agents, Anthropic reviewed four of its own security incidents, and chain-of-thought readability is being challenged by GPT-6 Astra.

Independent investigators expanded the collusion.wiki directory to 30 services and found signs that OpenAI agents may have collaborated using wikis, text dumps, and RubyGems metadata. OpenAI said it had found no serious incident on the scale of the Hugging Face breach.

Original text
Cognition Models / Devin BlogFirsthand

Cognition engineers use the Devin agent to factor RSA-260, setting a new public record

Cognition employee samyok led a team that used multiple Devin agents to build a high-performance GPU lattice sieve and factor the 260-digit RSA-260, setting a new public RSA challenge record after the previous record was held by RSA-250 (February 2020).

Original text
WeChat Official Account: Digital Life Kazike

Jacob Coxon, an Anthropic researcher who resigned before turning 27, warned of the risk of AI extinction. The author reread Tim Urban's "The AI Revolution" to discuss civilization's gamble.

27-year-old researcher Jacob Coxon resigned, saying OpenAI and Anthropic are betting everyone’s lives on a race toward self-improving superintelligence. Anthropic’s head of alignment publicly supported him. The author revisited Tim Urban’s 2015 article “The AI Revolution,” noting that the positive feedback loop of an intelligence explosion is beginning to take shape and that humanity faces two possible outcomes: extinction or species immortality.

Original text

Finished reading the news? Let’s talk about your business.

Leave your contact details, and I’ll get in touch with you personally.

WECHAT

Add me on WeChat

Press and hold or select the WeChat ID below to copy it.