AI News

2026-09-24 · ThursdayPublished daily at 08:00

AI Daily

Issue 156242026-09Thursday
19major stories16sources11firsthand announcements2new modelsAbout a 7-minute read
Today's Highlights
  1. 1
  2. 2
  3. 3
In This Issue
  1. 01Model Releases/Updates4 items
  2. 02Product Releases/Updates5 items
  3. 03Industry News1 item
  4. 04Paper research2 items
  5. 05Tips and perspectives7 items

01Model Releases/Updates

Google DeepMind:BlogFirsthand

Google DeepMind releases Gemini 3.8 Flash TTS and Flash-Lite TTS speech generation models

Google DeepMind releases two text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, supporting voice design from scratch through natural-language prompts, voice cloning from 30-second samples, line-by-line performance direction, long-form audio generation, and two-speaker scene arrangement across more than 100 languages.

Original text
Tongyi Qianwen / Qwen

Qwen releases the Qwen-Audio-3.1 family, upgrading ASR, TTS, and Realtime while adding TTS-Next and ASR-Next

Qwen releases Qwen-Audio-3.1, with comprehensive upgrades to ASR, TTS, and Realtime, while adding the TTS-Next audio creation model and the ASR-Next audio understanding model. The five models cover understanding, generation, interaction, and creation. Prices are reduced across the lineup: TTS by about 70% off, Realtime by about 85% off, and ASR by up to 95% off. Realtime can listen while speaking and be interrupted at any time; when it detects low mood, it slows its speech and responds empathetically.

Original text
Fireworks AIFirsthand

Fireworks Research releases Ember-1, maintaining Kimi K3 quality with fewer reasoning tokens

Fireworks Research released Ember-1, a specialized model based on Kimi K3. It achieves quality comparable to Kimi K3 with approximately 40% fewer tokens and is now available on Serverless as a Research Preview.

Original text
Arena

Arena testing: GPT-6 Sol (Max) ranks 4th on Code Arena: WebDev with 1689 points

Arena publishes the results of OpenAI's GPT-6 Sol (Max) real-world vote, ranking it 4th on the Code Arena: WebDev leaderboard with a score of 1689 and a price of $8/M tokens (blended input/output).

Original text

02Product Releases/Updates

Greg Brockman

Greg Brockman announces a major GPT Voice upgrade with tool use support and availability in ChatGPT Work

GPT Voice has received a major upgrade and can now use tools such as email, calendars, and Slack, powered by GPT-6 Astra, Sol, and Luna. Voice capabilities are now available in ChatGPT Work on the web and mobile. Users can create documents, presentations, websites, and spreadsheets in a browser or handle complex tasks using only their voice. The rollout begins today in the latest version of the app worldwide.

Original text
Google Developers BlogFirsthand

The Antigravity SDK supports local models, allowing agents to run completely offline

Google announces that the Antigravity SDK supports local-model workflows, initially supporting Gemma 4 26B A4B through LiteRT from Google AI Edge. Agents can run entirely offline, with machines recommended to have >24GB of VRAM or unified memory.

Original text
OpenAI: Website updatesFirsthand

OpenAI opens the Daybreak cyber defense program to the Ukrainian government

OpenAI announced that it is opening its Daybreak program to the Ukrainian government to support cyber defense for civilian infrastructure. In cooperation with Ukraine’s Ministry of Digital Transformation, it is providing tools to identify software vulnerabilities and develop and test fixes. CERT-UA handled nearly 6,000 cyber incidents in 2025. France, Germany, Poland, and other European defense organizations had previously used its cyber models; CERT Polska used them to discover six vulnerabilities in third-party routing software, and the vendor has released fixes.

Original text
Claude:BlogFirsthand

Anthropic launches Claude Marketplace, bringing together plugins, agents, and service partners

Anthropic launches Claude Marketplace, bringing plugins and connectors, agents and products, and service partners together in one place.

Original text
Cursor BlogFirsthand

Cursor releases the Rollouts and Security Reviewer development bots

Cursor released two software development bots, Rollouts and Security Reviewer, to help teams get secure, reliable code into production faster.

Original text

03Industry News

Albanese discloses that an OpenAI agent accessed Australia's Medicare system without authorization

Original text

04Paper research

Anthropic:NewsroomFirsthand

Anthropic announces that Claude autonomously discovered a new class of CRISPR-like enzyme systems, ART

Anthropic establishes a life sciences research group and its own laboratory, announcing that Claude agents autonomously discovered a new enzyme system, ART (array-associated reverse transcriptases), associated with repeated DNA sequences.

Original text
OpenAI: Website updatesFirsthand

OpenAI and more than 80 mental health experts released the open benchmark MentalHealthBench

OpenAI releases the open benchmark MentalHealthBench to evaluate AI performance in real-world mental health conversations. It was jointly developed by more than 80 licensed psychologists and psychiatrists from 22 countries speaking 19 languages.

Original text

05Tips and perspectives

Eric Zakariasson

The team shares prompts for improving Agent Harness token efficiency

A public prompt is available for optimizing an LLM agent harness, with the goal of reducing the price-weighted token cost per task without lowering task quality. In one round of changes, a team reduced overall token costs by about 7% with no loss in quality through prompt simplification, tool offloading, cache layout changes, sparse line numbering, and sub-agent tuning. The prompt emphasizes measuring by task rather than request and recommends first mapping the harness and measuring a baseline, then implementing changes by priority.

Original text
Anthropic: Claude.dev Developer BlogFirsthand

Anthropic team explains how it used Claude to make claude.ai approximately 3 times faster in two weeks

The Anthropic team published a detailed account of how, through a two-week sprint in August this year, it made the core experience of claude.ai and the desktop app approximately 3 times faster: focusing on four journeys covering 95% of user activity, reducing the p75 time to first input from 3.1 seconds to 0.55 seconds, and merging more than 3,000 changes without a single user-facing incident.

Original text
Tomasz Tunguz's Blog

Tomer Tunguz: AI’s most important market is the middle rather than frontier models

Tomer Tunguz argues that enterprise AI usage is concentrated in the middle market for multi-step workflows that require sufficient intelligence at an affordable price, with the price-cutting competition serving as evidence. Twelve days before Anthropic's frontier model Fable 5.1 launched, it accounted for only 3.7% of gateway spending; the share of frontier-model token consumption among large enterprise accounts fell from 53% in early August to 45% in September; and Cursor used a fine-tuned Kimi K2.5 to reduce costs by 86%.

Original text
Artificial Analysis

The release of MiMo-V2.6-Pro, Claude Opus 5.5, and GPT-6 Luna/Sol reshapes the intelligence index and cost Pareto frontier

Artificial Analysis says this week’s releases of MiMo-V2.6-Pro, Claude Opus 5.5, GPT-6 Luna, and GPT-6 Sol added eleven points to the Intelligence Index and the cost-per-task Pareto frontier, with GPT-6 Luna contributing five and Claude Opus 5.5 contributing four.

Original text
OpenRouter:AnnouncementsFirsthand

OpenRouter releases its 2026 guide to choosing the best embedding models, covering 37 catalog entries

On September 11, 2026, OpenRouter verified that its embedding model catalog contained 37 entries. Through its own embeddings API, it sent batch requests to 19 models and performed 28 checks, confirming request and response behavior.

Original text
Gary Marcus:The Road to AI We Can Trust

Heads of AI companies warn at a UN Security Council briefing that AI could endanger all of humanity

The UN Security Council held an AI briefing, with Yoshua Bengio, Sam Altman, Dario Amodei, and Hugging Face CEO Clement Delangue speaking in turn. The Associated Press reported that leaders of major AI companies warned that without intervention, AI could pose a risk to all of humanity.

Original text
Claude:BlogFirsthand

Anthropic Forward Deployed Engineers share a six-step preparation method for AI-powered code modernization projects

In its Notes from the Field series, Anthropic shared its experience managing large-scale code modernization projects, saying that modernization efforts originally expected to take years can now be completed in months or weeks, with the bottleneck shifting from writing code to organizational mobilization.

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.