- unwind ai
- Posts
- Google's Open Agent Payments Protocol
Google's Open Agent Payments Protocol
+ Alibaba's open-source 3B agentic model beats Grok-4 and GPT-5
Shubham Saboo & Gargi Gupta
September 17, 2025
In partnership with
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
Learn OpenAI Agents SDK from zero to production-ready!
We have created a comprehensive crash course that takes you through 11 hands-on tutorials covering everything from basic agent creation to advanced multi-agent workflows using OpenAI Agents SDK.
What you'll learn and build:
Starter agents with structured outputs using Pydantic
Tool-integrated agents with custom functions and built-in capabilities
Multi-agent systems with handoffs and delegation
Production-ready agents with tracing, guardrails, and sessions
Voice agents with real-time conversation capabilities
Each tutorial includes working code, interactive web interfaces, and real-world examples.
The course covers the complete agent development lifecycle: orchestration, tool integration, memory management, and deployment strategies.
Everything is 100% open-source.
We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Latest Developments
A tiny 3B parameter model just outperformed xAI Grok 4, GPT-5, Gemini 2.5 Pro, and Deep Research tools like Kimi Researcher on multiple research benchmarks, and it's completely open-source.
Alibaba's Tongyi Deep Research proves that smart architecture beats raw scale, delivering state-of-the-art web research capabilities while using a fraction of the compute.
Built with 30B total parameters but only 3B activated per token, Tongyi DeepResearch is specifically designed for long-horizon, deep information-seeking tasks, scoring 32.9 on Humanity's Last Exam.
The model comes with two inference modes: ReAct for standard operation and Heavy Mode which uses parallel research agents to push performance even higher. What makes this release special is the complete package - Alibaba open-sourced not just the model weights but the entire training pipeline, synthetic data generation framework, and reinforcement learning approach that made these results possible.
Key Highlights:
Efficiency Champion - Achieves superior performance to OpenAI and Kimi Deep Research across multiple benchmarks while using only 3B activated parameters.
Zero Human Annotation - Generates all training data automatically through multi-stage synthesis pipeline, creating both diverse interaction data and "super-human" difficulty scenarios without manual labeling costs.
Dual Inference Modes - Offers both ReAct mode for standard performance and Heavy Mode that uses parallel research agents with IterResearch paradigm, allowing you to choose between speed and maximum capability.
Complete Open Research - The team has open-sourced model weights on Hugging Face and Modelscope, along with training code, RL framework, and evaluation tools, all under Apache 2.0 license.
AI agents can finally handle your credit card without you worrying about unauthorized purchases or sketchy transactions.
Google just dropped Agent Payments Protocol (AP2), an open protocol to securely initiate and transact AI agent-led payments across platforms.
Current payment systems assume a human is clicking "buy now," but when your shopping agent tries to grab those concert tickets, banks have no clue if you actually authorized it or if the AI just hallucinated the whole thing.
AP2 introduces cryptographic "verifiable credentials" that create an unbreakable audit trail every step of the transaction - your agent gets a digitally signed mandate from you, the merchant signs off on the exact cart details, and payment networks can verify everything actually happened as intended.
It’s designed to work alongside MCP and A2A protocols where AI agents use external tools, discover and interact with other AI agents, and safely execute payments on your behalf.
Key Highlights:
Role-based Architecture - AP2 has an actor/agent at every stage, where each has a specific role to play. There are User Agents (acting on behalf of the user), Credential Providers (payment security), Merchant Endpoints (product catalog), and Payment Processors (transaction execution) to prevent any single point of failure.
Cryptography - Each transaction generates tamper-evident digital objects that create an immutable audit trail, making it nearly impossible for agents to spend money without proper authorization or for disputes to go unresolved.
Progressive Autonomy - Currently, AP2 supports scenarios where a user is present to specifically authorize a transaction that the AI agents can execute. Future versions of AP2 will also support events where humans are not actively present - they just authorize the agent to make a payment under specific conditions. For example, buying an item if it goes under $100.
Backwards-Compatible - Works with existing EMV and credit card infrastructure while adding support for emerging payment methods like real-time bank transfers and digital currencies across different countries.
How 433 Investors Unlocked 400X Return Potential
Institutional investors back startups to unlock outsized returns. Regular investors have to wait. But not anymore. Thanks to regulatory updates, some companies are doing things differently.
Take Revolut. In 2016, 433 regular people invested an average of $2,730. Today? They got a 400X buyout offer from the company, as Revolut’s valuation increased 89,900% in the same timeframe.
Founded by a former Zillow exec, Pacaso’s co-ownership tech reshapes the $1.3T vacation home market. They’ve earned $110M+ in gross profit to date, including 41% YoY growth in 2024 alone. They even reserved the Nasdaq ticker PCSO.
The same institutional investors behind Uber, Venmo, and eBay backed Pacaso. And you can join them. But not for long. Pacaso’s investment opportunity ends September 18.
Paid advertisement for Pacaso’s Regulation A offering. Read the offering circular at invest.pacaso.com. Reserving a ticker symbol is not a guarantee that the company will go public. Listing on the NASDAQ is subject to approvals.
Quick Bites
Build a v0-like vibe coding app with this open-source kit
Vercel just dropped v0-clone, an open-source starter kit that lets you build your own AI app builder using their v0 SDK. The kit comes loaded with everything you'd expect - real-time streaming, multi-tenant auth, conversation history, and drag-and-drop attachments - essentially giving you a full v0.dev replica out of the box. You can spin it up with a single command and have your own AI app builder running in minutes.
Gamma releases AI design partner to go beyond just slide decks
Gamma shipped 3.0 with two standout features that push AI presentation tools into new territory. The Gamma Agent can now make sweeping edits across entire decks with a single prompt. Tell it to "make it more visual" and it scans each slide for data worth charting, or upload rough notes and watch it synthesize, research, and format everything to match your brand. Along with this, the new Gamma API lets you automate deck creation at scale. Connect to Zapier or other workflows to spin out presentations from meeting transcripts or data sources.
GitHub releases MCP Registry for easy discovery and installation
GitHub launched an MCP Registry with 38 curated MCP servers, adding to the growing pile of MCP directories already out there. You can install these MCP servers in VS Code in just one click, which is useful. But developers need something else: tools to actually test and evaluate servers before committing, rather than another list. The real challenge isn't discovery - it's figuring out which of the thousands of available MCP servers actually work reliably with AI agents without broken implementations. Hope GitHub tries solving this.
Free AI code reviews in your CLI with CodeRabbit
CodeRabbit has released CodeRabbit CLI, an AI code review agent that analyzes your code directly in the terminal, catching bugs and enforcing standards before commits. Consider it as a quality gate that complements other CLI coding agents like Claude Code, Gemini CLI, and Codex CLI. You can ask these agents to use CodeRabbit and fix any issues it finds. CodeRabbit CLI is free to use with rate limits.
Tools of the Trade
Orchids - Full-stack AI Engineer that lets you vibe code applications complete with frontend, backend, databases, APIs, auth, payments, and everything else without manually configuring any third-party apps. It ranks #1 on UI Bench and Design Arena - beating Devin, Lovable, Cursor, Bolt, Replit, and v0.
AI Apply - AI that searches 100s of career websites and applies autonomously on your behalf, based on your specified criteria. It doesn’t just apply; it also optimizes your resume and cover letter as per the job description for better chances of getting noticed.
Riveter - Manual research to fill spreadsheets sucks. Riveter deploys multiple AI agents to automate research across spreadsheet rows, with each cell getting its own agent to gather data from websites, PDFs, and images.
Awesome LLM Apps: A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
(Now accepting GitHub sponsorships)
Hot Takes
That’s all for today! See you tomorrow with more such AI-filled content.
Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
Reply