Skip to main content
Back

OpenAI unveils GPT-5.2, reaches human expert-level work

On Cursor's Visual Editor, Google's Research Agent, Disney-OpenAI Deal, Claude Code in Android, Anthropic's 2026 Program

December 13, 2025 3 min read

Hey,

Your daily briefing is ready. You can finally take a break from the AI firehose.

Our algos spent the night splitting signal from noise and pulled the top news, models, papers, and repos.

Here’s the must-read:

Summary

Read time: 3 min 36 sec

Top News

▸ OpenAI launches GPT-5.2 to improve professional workflows and agent reliability

Top News

▸ Cursor adds a visual editor that writes real code from direct UI edits

Top News

▸ Google debuts Gemini Deep Research, an agent for multi-step web research

Signals

▸ Disney invests $1B as OpenAI licenses characters for Sora content
▸ Google releases Disco to turn open browser tabs into custom AI apps
▸ Anthropic brings Claude Code to Android with improved model switching
▸ Thinking Machines makes Tinker available to all, its open-source fine-tuning API
▸ Anthropic opens 2026 Fellows Program for funded, mentored AI safety research
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Top News
OpenAI releases GPT-5.2, boosting reasoning, coding reliability, and long-context performance for production agents
42,759 Likes
Grok 4 Fast Benchmark

OpenAI introduces GPT-5.2 as a model family built for real work, not demos. Teams already rely on GPT-5.1 for code, docs, and agents, but limits show up fast with long context, fragile tool chains, and repo-scale code.

GPT-5.2 breaks that ceiling by reaching human-expert performance on real knowledge work tasks.

The release follows GPT-5.1 and responds to growing competition in long-context, coding, and agent workflows. The most notable result is GPT-5.2 Thinking reaching human-expert performance on GDPval knowledge work tasks.

What GPT-5.2 adds

  • 70.9% win-or-tie rate versus industry professionals on GDPval knowledge work tasks.

  • 55.6% on SWE-Bench Pro and 80.0% on SWE-Bench Verified repo-level coding.

  • Near 100% accuracy on MRCRv2 long-context tests at 256k tokens.

  • 98.7% accuracy on τ2-bench Telecom multi-tool workflows.

How you use it

  • Select Instant for fast answers and writing.

  • Select Thinking for codebases, spreadsheets, and long documents.

  • Select Pro when accuracy matters more than latency.

  • Use the API with 'gpt-5.2' and raise reasoning effort to 'xhigh' when needed.

Try now
Top News
Cursor introduces a browser-based visual editor for drag-and-drop UI changes mapped to code
12,001 Likes
Grok 4 Fast Benchmark

Cursor adds a visual editor to the Cursor Browser that lets you change a live web app and update its code at the same time. Frontend work often splits into two loops: tweak UI visually, then hunt through files to recreate the change in code.

Cursor targets that gap by letting you act directly on the rendered interface and delegate code edits to an agent.

The key insight is simple: the browser already knows what you see, and the codebase already knows how it renders. Cursor connects the two.

You grab elements on the page, adjust them, or describe changes in text. The agent traces those actions back to the right components and rewrites the source files.

The editor works on real React applications, not mockups. It exposes component props so you can test different states, like loading or error views, without manual edits.

Key features and behavior

  • Drag-and-drop reorders elements directly across the DOM tree

  • Sidebar exposes React component props for state and variant testing

  • Visual controls adjust colors, typography, grids, and flexbox layouts

  • Point-and-prompt applies text instructions to selected elements

  • Multiple agents run in parallel and apply changes within seconds

Try now
Top News
Google unveils a research agent that autonomously searches, verifies facts, and generates structured, cited findings
3,799 Likes
Grok 4 Fast Benchmark

Google introduces the Gemini Deep Research agent at a moment when developers already scrape, search, and stitch answers by hand. Research-heavy apps rely on brittle pipelines that stop after one query or miss sources.

Google’s insight is simple: treat research as an iterative process, not a single request.

Gemini Deep Research runs on Gemini 3 Pro and acts like a self-directed researcher. It plans an investigation, searches the web, reads results, spots gaps, and searches again.

This loop continues until it produces a structured report with citations tied to each claim. Reinforcement learning trains the agent to favor accuracy and reduce unsupported statements.

The breakthrough shows up in benchmarks and tooling. Google exposes the agent through the new Interactions API, which lets you call models, agents, and tools from one endpoint.

You can upload documents, mix them with web sources, and control report structure through prompts.

Key capabilities and results

  • Scores 46.4% on Humanity’s Last Exam, testing combined reasoning and factual knowledge

  • Achieves 66.1% on DeepSearchQA, a new 900-task multi-step web research benchmark

  • Uses autonomous planning, query iteration, and gap detection during execution

  • Returns claim-level citations and structured JSON outputs

  • Accepts PDFs, CSVs, and documents alongside live web data

Get started now
Signals
1 OpenAI partners with Disney to power fan-prompted Sora videos while committing to responsible AI safeguards 154,793 Likes
2 Google introduces Disco, letting Gemini 3 transform your open browser tabs into custom web apps 5,421 Likes
3 Anthropic expands Claude Code with Android support, faster model switching, and detailed context window indicators 3,281 Likes
4 Thinking Machines rolls out Tinker generally, adding Kimi K2 Thinking, vision input, and OpenAI-compatible sampling 1,739 Likes
5 Anthropic announces 2026 Fellows Program applications, funding four-month AI safety and security research projects 1,711 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us