Welcome back!

Claude is stepping up its game. Finally, Anthropic’s AI assistant can search the web through a Brave partnership, providing real-time information with direct citations for more trustworthy answers. This levels the playing field with ChatGPT. Let’s unpack…

Today’s Summary:

• 🚀 Claude launches web search

• 💻 Nvidia unveils AI supercomputers & robotics model

• 🗣️ OpenAI sets new voice accuracy record

• 📖 OLMo 2 32B fully open-source model released

• 🇪🇺 Meta’s AI assistant goes live in the EU

• 🖼️ xAI launches API for image generation

• 🛠️ 2 new tools

TOP STORY

Claude can now search the web via Brave partnership

The Summary: Anthropic has launched web search for Claude, giving it access to real-time information from the web. This closes a key competitive gap with ChatGPT. Claude Search provides direct citations to sources, helping users verify information accuracy. Developers have found evidence in Claude’s internal prompt parameters suggesting the search is provided by Brave Search.

Key details:

• Web search requires activation in Profile Settings and is currently limited to paid US users on Claude 3.7 Sonnet

• Enterprise and Team plans need workspace-level activation by an Owner before individual users can toggle the feature on

• When searching, Claude shows an indicator, processes multiple sources, and provides conversational responses with citations

• Technical analysis reveals "BraveSearchParams" in Claude's code and identical search results between Claude and Brave Search

Why it matters: 

Claude has lagged behind ChatGPT in search, but the high quality of Claude 3.7 Sonnet responses could make up for the integration delay. Unlike basic search aggregation, Claude can synthesize information and reason through complex queries, creating a more sophisticated user experience.

NVIDIA

Nvidia unveils multiple AI breakthroughs at GTC

The Summary: Nvidia introduced several AI breakthroughs at GTC 2025, including personal AI supercomputers DGX Spark and DGX Station. The company also introduced Cosmos-Transfer1 for robot training, and GR00T N1, the first open foundation model for humanoid robots with a dual "fast and slow" thinking system. These innovations aim to transform robotics development and bring data center-level AI computing to desktops.

Key details:

• DGX Spark (formerly Project DIGITS) is a personal AI supercomputer capable of 1,000 trillion operations per second in a desktop-sized unit

• GR00T N1 robotics AI features a two-part brain: "System 1" for fast reflexes and "System 2" for methodical reasoning

• NVIDIA, Google DeepMind, and Disney Research are jointly developing Newton, an open-source physics engine that accelerates robotics workloads by 70x

Why it matters: By combining high-performance chips, open-source models, and physics-based simulation, Nvidia is building an AI stack that spans from data centers to physical robots and personal supercomputers. This strategy positions Nvidia not just as a hardware supplier, but as a foundation for how AI is trained, deployed, and integrated into real-world systems.

GTC 2025 Full Video Keynote

OPENAI

OpenAI Voice Models set new accuracy records

The Summary: OpenAI has released three new audio models that deliver cutting-edge voice capabilities to developers. The gpt-4o-transcribe speech-to-text model sets a new benchmark with a 2.46% word error rate in English, while its mini version offers a cost-effective alternative. For the first time, the gpt-4o-mini-tts text-to-speech model lets developers control how it speaks, such as “talk like a medieval knight” or “talk like a sympathetic customer service agent”.

Key details:

• New models achieve 30% improved transcription accuracy

• Voice AI can be customized via text prompts ("speak like a mad scientist" or "talk like a mindfulness teacher")

• Integration requires just 9 lines of code to add fluid voice capabilities to existing GPT-4o applications

• Pricing starts at $0.003 per minute for transcription and $0.015 per minute for text-to-speech

Why it matters: These models make it easy for developers to add voice to existing text applications with minimal code changes, putting OpenAI at the forefront of the rapidly evolving voice AI landscape. The ability to create emotionally responsive, customizable voice agents opens new possibilities to enhance user experiences while addressing previous “voice imitation” concerns.

API documentation

QUICK NEWS

Quick news

• OLMo 2 32B fully open source model with published training data

• xAI launches an API for image generation

• Meta launches its AI assistant in EU with text-only capabilities

TOOLS

🥇 New tools

• Touch Grass - Reduce your screen time

• TalkMe - Personal language tutor