Google AI launches Gemini 3.8 Flash and Flash-Lite TTS

#1 𝕏

Google AI released Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, supporting custom voices across 100+ languages and 2,000+ ready-to-use voices. Flash targets high-fidelity creative production, while Flash-Lite is designed for near-real-time voice agents, high-volume dubbing, and bulk audio creation.

Also covered by: @Google DeepMind, @Philipp Schmid, @Philipp Schmid, @Google DeepMind, @Google AI, @Logan Kilpatrick, @Logan Kilpatrick

#2 𝕏

Cursor announced Rollouts, which write a monitoring plan and watch changes as they deploy. Deployments are verified to catch regressions before users see them.

#3 𝕏

New ChatGPT Voice capabilities include plugins for email, calendar, and Slack; support for GPT-6 Astra, Sol, and Luna; and voice-driven creation of documents, presentations, sites, and spreadsheets, as well as handling complex browser tasks, in ChatGPT Work on web and mobile. The capabilities are rolling out globally in the latest app version.

#4 📝 Anthropic News

Claude discovers a novel enzyme system with CRISPR-like repeats - Anthropic’s new life‑sciences lab used Claude to search DNA databases with roughly 950 agents over 21 hours consuming about 210 million tokens and autonomously identified a previously uncharacterized enzyme system called array‑associated reverse transcriptases (ART)—a jumbo‑phage reverse transcriptase found next to arrays of non‑coding DNA and an accessory protein reminiscent of CRISPR—which they began validating in the lab and reported in a preprint.

Also covered by: @Anthropic, @Dario Amodei

#5 𝕏

Qwen announced Qwen-Audio-3.1, a five-model stack with upgraded ASR, TTS and Realtime plus new ASR-Next and TTS-Next models for audio understanding and creation. The post claims price cuts of TTS ~70% off, Realtime ~85% off and ASR up to 95% off.

#6 𝕏

Qwen released Qwen Intelligence with three “SOTA” agents for task planning, API-first mobile execution with GUI fallback, and one-sentence creative generation. Qwen reports a #1 Planner ranking, Mobile-Use scores of 82.1/92.2/97.2 and 90% end-to-end success, three-second image generation at about 2x peer speed, and opened four benchmarks covering planning, execution, real-device performance, and safety.

#7 📝 OpenAI News

ChatGPT Ads expands to Southeast Asia and Taiwan - ChatGPT Ads is rolling out across Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, and Taiwan, expanding availability to more than 60 countries after earlier APAC launches in Australia, New Zealand, Japan, South Korea, and India. Ads are shown only to Free and Go users (Pro and Enterprise remain ad-free), advertisers can buy through OpenAI’s Ads Solutions team, agency partners (dentsu, Havas Media, Omnicom Media, Publicis Groupe, WPP) or self-service Ads Manager, and OpenAI reports ChatGPT Ads hit a $1 billion annualized revenue run rate in under 200 days with tens of thousands of advertisers.

Also covered by: @Udi Menkes

#8 𝕏

Boris Cherny shared a blog post explaining how the claude.ai login experience and Desktop app became faster in the last few weeks, with techniques for engineers speeding up their own apps.

#9 𝕏

Boris Cherny commented that Claude appears to model tricky state machines or race-prone code, identify counter-examples as suspected bugs, reproduce them, and fix the code. He clarified that this targets the hardest parts rather than formally verifying the entire codebase.

#10 𝕏

NVIDIA AI released the 100M-parameter Nemotron 3 Diarization model, which tracks who spoke when—even with overlapping voices—and handles up to eight speakers. The model is available on Hugging Face.

Also covered by: @NVIDIA AI

#11 𝕏

Cognition announced two updates to Devin for Microsoft users: native Microsoft Teams support and a first-party Microsoft 365 integration connecting to mail, calendar files, chats, and more.

Also covered by: @Cognition

#12 𝕏

OpenAI released MentalHealthBench, a new open benchmark built with input from more than 80 mental health clinicians, to demonstrate frontier-model improvements in realistic mental health conversations and enable researchers to examine its methods, run evaluations, and build on the work.

#13 𝕏

Cursor shared a link explaining how it improved its agent harness.

#14 ▶️

Building AI Assistants with On-Device Memory

Deeplearning.ai

Also covered by: @DeepLearning.AI

#15 𝕏

Julien Chaumond announced the release of the huggingface/lerobot JavaScript package, which reads @LeRobotHF datasets on the Hub directly in the browser without downloads. It supports custom dataset viewers, including an example UI built in ~600 lines of JS, with its GitHub repository linked in a reply.

#16 in

Guillermo Rauch announced Drives, an early storage companion to Sandbox that acts as an attachable external disk, enabling file access without booting an agent’s full computer. He argues that separating agents into 3 key components—brain, hands, and files—can reduce costs while improving security and auditability.

#17 𝕏

bolt.new recapped one week of open-model usage on Bolt Forge: GLM 5.3 Flash accounted for 63% of prompts and DeepSeek V4.1 Flash for 17%, totaling 80%. Bigger models were available, but builders repeatedly chose faster options; the linked page listed a price of $9/month.

#18 𝕏

Garry Tan said he uses @capydotai daily on all his GStack/GBrain pull requests and is going at least 4x faster than with raw Codex or Claude Code via Conductor. He added that it handled something on a recent pull request that out-of-the-box Codex would not have done, though the specific action was not provided.

#19 𝕏

Anthropic shared that @CEPIvaccines, @WHOAFRO, and @inrb_kinshasa are using Claude to accelerate their response to an outbreak of an unusual Ebola variant in the Democratic Republic of the Congo.

#20 𝕏

Santiago shared an article about converged databases and thanked the Oracle team for partnering with him on the post.