ElevenLabs Now Generates Studio-Quality Music🎶🎼✨

AI innovation is breaking boundaries across creative, cognitive, and immersive domains.

ElevenLabs just launched Eleven Music, a text-to-music generator that composes full studio-quality tracks—vocals, genres, lyrics, and transitions—instantly.

Meanwhile, OpenAI unveiled gpt-oss-120b and gpt-oss-20b, two powerful open-weight language models optimized for performance and customization, pushing open-source AI to the next level.

And Google DeepMind introduced Genie 3, a real-time world model capable of turning text into interactive, navigable environments—perfect for research, games, and generative media.

Let’s dive deeper into these groundbreaking AI developments. 👇️ 

ElevenLabs Rolls Out AI Music Platform for Fast, Customized Song Creation

ElevenLabs has introduced Eleven Music, an advanced AI music generator that enables users to quickly create studio-quality tracks in any genre or style using simple text prompts—including options for vocals or instrumentals and support for multiple languages. Designed for businesses, creators, and music lovers, the platform allows full customization—users can generate, edit, and structure songs section by section, control mood shifts and transitions, and fine-tune duration, lyrics, and musical style. Eleven Music’s proprietary AI model delivers real-time, high-fidelity compositions, blending genres and instruments seamlessly for both commercial and creative uses, such as cinematic scores, ad jingles, games, and podcasts. The platform also offers an API (coming soon) for programmatic music generation and integration into other products or workflows, with usage rights available for a variety of commercial applications.

GPT OSS Powerful Open-Weight Language Models Released

OpenAI has released gpt-oss-120b and gpt-oss-20b, two cutting-edge open-weight language models under the Apache 2.0 license. These models surpass similarly sized open models in reasoning and tool use, are optimized for efficiency on consumer hardware, and support extensive customization. gpt-oss-120b features 117 billion parameters and rivals OpenAI’s o4-mini in reasoning benchmarks, running on a single 80GB GPU, while the 21 billion-parameter gpt-oss-20b matches o3-mini and runs on devices with just 16GB of memory. Both models use a Mixture-of-Experts (MoE) architecture, support up to 128,000 token context lengths, and are evaluated to meet strong safety standards. OpenAI collaborated with industry partners for real-world application, rigorously tested the models’ safety—especially against adversarial fine-tuning—and made the weights, tokenizer, and tools openly available for researchers and developers to run and fine-tune on their own infrastructure.

Google DeepMind Launches Genie 3: Real-Time Interactive World Model

Genie 3, announced by Google DeepMind, is a groundbreaking general-purpose world model capable of generating a remarkably wide variety of interactive environments in real time from simple text prompts. It enables users to explore dynamic, consistent, and visually immersive worlds at 24 frames per second and 720p resolution—retaining physical and environmental continuity for a few minutes. Genie 3 models natural and fantastical phenomena, supports user-driven navigation and “promptable world events,” and demonstrates emergent abilities like maintaining scene consistency over longer periods and recalling locations visited earlier. Designed as a foundational AI tool for research, agent training, and generative media, Genie 3 is currently offered as a limited research preview to select academics and creators, with future plans for broader access. While the model excels in versatility and realism, it still faces limits such as a constrained action space, challenges with long-duration interactions, and imperfect representations of real-world locations. Google DeepMind’s approach emphasizes responsibility and community collaboration as they continue to refine this new class of world model technology.

Hand Picked Video

In this video, we’ll look at how to remove and replace video backgrounds using our new tool—no green screen, no complex editing—just upload and go.

Top AI Products from this week

• Asteroid - Asteroid lets anyone build highly custom, complex AI browser agents. Both non-technical users and engineers can now reliably automate back office browser tasks 20x faster with Asteroid’s web-based agent builder, API integration, and more.

• Indy AI by Contra - Contra is the commission-free creative network, connecting you with the talent and tools to get work underway. Hire more independents. Start more projects. Get more creative.

• Embeddable - Build interactive tools like forms, popups, quizzes, and scratch cards to boost engagement and convert more leads. No code needed. Just type a prompt, customize the result, and embed a lightweight, SEO-friendly script on your site in minutes.

• Voice Agents by Perspective AI - Your customers have a lot to say. Let them say it. Voice Agents lets your customers talk naturally—anytime, anywhere, in any language. Researchers say it's the fastest way to get honest insights, and they've never felt more connected to their customers.

• Writingmate 3.0 - Save money with all in one AI subscription. Chat, code, compare models, search the web and generate images using best LLMs in one place.

• Flowtica Scribe - Flowtica Scribe is an AI-powered pen that records audio and lets you highlight key moments with a simple button press. It then generates structured, personalized notes focused on what you decided was important.

This week in AI

• Character.AI Launches AI-Native Social Feed - Character.AI unveils the world’s first AI-native social feed, blending creation and consumption with interactive, remixable content on its mobile app.

• Qwen-Image: 20B MMDiT Model - Qwen-Image is a 20 billion parameter MMDiT model excelling in complex bilingual text rendering and precise image editing, supporting diverse fonts, multi-line layouts, and photorealistic to anime styles.

• OpenAI Removes Chats from Google - OpenAI pulled a feature that unintentionally exposed private ChatGPT chats in Google search results after user backlash over sensitive content visibility.

• Perplexity Uses Stealth Crawlers - Perplexity evades site blocks by disguising crawlers with undeclared user agents and IPs, ignoring robots.txt, prompting Cloudflare to block its stealth crawling activity.

• Google’s New Kaggle AI Game Arena - Google's new Kaggle Game Arena benchmarks AI by hosting head-to-head matches in strategic games like chess, aiming to reveal real-time reasoning and planning abilities.

Paper of The Day

EDCIM (Error Detection and Correction for Interpretable Mathematics) enhances large language models' math problem-solving by generating equations, detecting errors with symbolic analysis, and selectively correcting mistakes before final solution computation. This approach boosts accuracy, reduces costs, and improves partially wrong solutions, offering an efficient, interpretable pipeline for math and similar structured tasks.

To read the whole paper, go to here.