Cursor Composer undercuts competition: All the top news from Google’s I/O event

In today’s edition of Data Points, you’ll learn more about:

• Google’s remade Antigravity, an alternative to IDEs
• Omni Flash, Gemini’s voice-to-video generator
• Google’s AI overhaul of web search
• Corti’s Symphony, a specialist in medical transcription

But first:

Cursor’s coding model rivals leaders at lower price

Cursor shipped Composer 2.5, a coding model built on Moonshot’s open-source Kimi K2.5 and trained on 25 times more synthetic tasks than its predecessor. It scores 79.8 percent on SWE-Bench Multilingual,  beating GPT-5.5 (77.8 percent) and coming within one point of Claude Opus 4.7 (80.5 percent), and 63.2 percent on CursorBench v3.1, broadly in line with both frontier models. It's not clear whether Cursor has achieved true parity: Comparisons mix Cursor’s own harness with self-reported competitor numbers and have not yet been independently reproduced on a unified scaffold. But Composer operates at a fraction of the cost: $0.50/$2.50 per million input/output tokens versus Anthropic and OpenAI’s substantially higher rates. A faster variant delivers the same performance at $3.00/$15.00 per million tokens.  (The Decoder)

Google updates speedy Flash model, raises prices

Google introduced Gemini 3.5 Flash, built to power autonomous agents at frontier-level intelligence. It outperforms Gemini 3.1 Pro on coding and agentic benchmarks, runs four times faster than competing frontier models, and makes long-horizon tasks that previously took days or weeks practical. 3.5 Flash is available immediately across the Gemini app, Google Search, the Gemini API, and enterprise platforms; a more capable 3.5 Pro variant arrives next month. Early partners including Shopify, Macquarie Bank, and Salesforce are already deploying it for complex workflows (financial document processing, merchant forecasting) at less than half the cost of alternative frontier models. The release also powers Gemini Spark, a new personal AI agent designed to take autonomous action on behalf of users throughout their day. (Google)

Rethinking (or replacing) the IDE with Antigravity

Google launched Antigravity 2.0, a standalone desktop application built around agents rather than traditional IDE concepts, available on macOS, Linux, and Windows. The move separates the agent-optimized interface from the Antigravity IDE, acknowledging that agent-first workflows now extend beyond coding into broader knowledge work. The new application introduces dynamic subagents that parallelize subtasks without bloating context windows, asynchronous task management, cron-style scheduled tasks, and slash commands like /goal (run to completion without asking) and /grill-me (ask clarifying questions before implementing). Voice input transcribes live rather than collecting raw audio, and conversations are grouped by “project” instead of “repository,” giving agents broader context while enforcing granular guardrails. Gemini’s popular CLI tool will be retired in favor of a new Antigravity CLI. The move reflects a deliberate judgment that agents have matured beyond coding acceleration, and that bundling IDE and agent interfaces confuses users unfamiliar with development tools. (Google)

Omni Flash, a new any-to-any multimedia generator

Google introduced Gemini Omni Flash, a multimodal model that generates and edits video through natural language. It accepts images, audio, video, and text, maintains character consistency and physics across multiple editing turns, and draws on real-world knowledge (physics, history, science) to move beyond pattern-matching into recognizable storytelling. The model launches today for Google AI Plus, Pro, and Ultra subscribers via the Gemini app and Google Flow, with free access rolling out to YouTube Shorts and YouTube Create. All videos carry Google’s SynthID digital watermark. (Google)

Google overhauls dominant search engine in favor of generated responses

Google rolled out a redesigned search interface that replaces separate AI Overviews and AI Mode with a unified AI Search experience built on Gemini 3.5. The new search box handles text, images, files, and video in a single query, supports follow-up questions, and builds context across a conversation for more personalized responses. Available immediately on desktop and mobile worldwide. (Google AI on X)

Specialized medical transcription model beats frontier AI

Copenhagen-based Corti launched Symphony for Speech-to-Text, a clinical-grade recognition model that achieved a 1.4 percent word error rate on English medical terminology—versus OpenAI’s 17.7 percent, ElevenLabs’ 18.1 percent, Whisper’s 17.4 percent, and Parakeet’s 18.9 percent. The gap widens further on structured clinical entities like medication dosages: Corti hit 98.3 percent recall while the strongest generalist model managed 44.3 percent. That difference matters more now than it used to. As healthcare shifts toward autonomous AI agents making real-time clinical decisions, transcription errors compound—if a model mishears “hyperthyroidism” as “hypothyroidism,” every downstream system operates on corrupted data. Corti also outperformed legacy incumbent Dragon Medical One in dictation accuracy and now serves over 100 million patients annually across health systems including the UK’s National Health Service. (VentureBeat)

Want to know more about what matters in AI right now? 

Read the latest issue of The Batch for in-depth analysis of news and research.

Last week, Andrew talked about the development of AI Andrew, an AI companion shaped by his personality, and his beliefs about respectful, empathetic, and technically precise communication.

“Many people are trying to understand what AI means for their work, learning, and careers. I’ve frequently enjoyed conversations with people on these topics. If you’d like to have a conversation on this, you might find AI Andrew can be a helpful thought partner and maybe even a friend — someone you can speak with about AI concepts, project ideas, career decisions, and whatever else is on your mind.” 

Read Andrew’s letter here.

Other top AI news and research stories covered in depth:

• The U.S. government plans to evaluate upcoming AI models for national security risks and other hazards prior to their release.
• OpenAI is challenging leaders in speech-to-speech technology with RealTime API updates that enhance audio models' reasoning, transcription, and translation capabilities.
• China's state regulators have blocked the Meta-Manus acquisition, preventing the tech giant from acquiring the agentic startup based in Singapore.
• Google's breast cancer detection models are being tested under real-world conditions in clinics, as two studies assess their effectiveness in mammogram diagnosis.

A special offer for our community

In case you missed it, DeepLearning.AI launched our first-ever subscription plan for our entire course catalog! As a Pro Member, you’ll immediately enjoy access to:

• Nearly 200 AI short and long courses from Andrew Ng and industry experts
• Labs and quizzes to test your knowledge
• Projects to share with employers
• Certificates to testify to your new skills
• A community to help you advance at the speed of AI

Enroll now to lock in a year of full access for $25 per month paid upfront, or opt for month-to-month payments at just $30 per month. Both payment options begin with a one-week free trial. Explore Pro’s benefits and start building today!

Try Pro Now

Data Points is produced by human editors with AI assistance.