OpenAI rival launches agent army

Good morning, AI enthusiasts. Manus AI just unveiled Wide Research, enabling users to deploy hundreds of AI agents simultaneously to tackle research tasks that previously required teams working for days.

With supercomputing-level research power now accessible through simple conversation, are we witnessing the democratization of institutional-grade analysis?

Friendly reminder: Our newsletter is still slowly moving to a new email address. Add [email protected] to your contacts to ensure you keep receiving our emails.

In today’s recap:

• Manus deploys agent swarms for parallel research

• Universal API unlocks all frontier AI models

• Open-source self-improving AI emerges

• AI scientist beats humans at toughest science tests

MANUS AI

Manus AI massively speeds up research tasks

Recaply: OpenAI rival Manus just unveils Wide Research, a new feature that lets users tackle large-scale research tasks by running hundreds of AI agents simultaneously in the cloud.

Key details:

• Multiple AI agents work in parallel to research hundreds of items at once.

• Handles complex tasks like analyzing Fortune 500 companies or comparing MBA programs.

• Results delivered in structured, sortable formats instead of endless browser tabs.

• Built on cloud-based virtual machines for maximum computing power.

• Available now for Pro users, rolling out to Plus and Basic tiers soon.

• Goes beyond research to enable advanced parallel processing workflows.

Why it matters: Wide Research democratizes supercomputing power through simple conversation. Instead of spending days manually researching hundreds of data points, users can now get comprehensive analysis in minutes. This represents a major shift toward AI agents handling complex, time-intensive tasks that previously required teams of researchers.

POE

Poe's new all-in-one API

Recaply: Quora Poe just introduced a new API that gives developers access to every AI model on their platform through a single interface, making it easier to build apps with multiple AI providers.

Key details:

• Works with existing Poe subscriptions - no separate billing needed.

• Compatible with popular developer tools like Cursor, Cline, and Continue.

• Includes text, image, video, and audio models from OpenAI, Anthropic, Google, and others.

• Uses standard OpenAI chat completion format for easy integration.

• Switch between models like Claude, GPT, or Grok with one parameter change.

• Can purchase additional points beyond subscription limits at cost.

Why it matters: This simplifies AI development by removing the need to manage multiple API keys and billing accounts. Developers can now access dozens of frontier models through one unified interface, potentially speeding up AI application development and reducing integration complexity.

DEEP COGITO

Open-source AI models learn to upgrade themselves

Recaply: Deep Cogito just released four powerful open-source AI models that can actually improve themselves, marking a major breakthrough in artificial intelligence development.

Key details:

• Four Cogito v2 models released (70B, 109B MoE, 405B, 671B MoE) with open licensing.

• Built using Iterated Distillation and Amplification (IDA) - similar to techniques used in AlphaGo.

• Models develop stronger "intuition" instead of relying on longer search processes.

• 671B model performs on par with latest DeepSeek models and approaches frontier closed models.

• Remarkably efficient training at under $3.5M total cost across all model sizes.

• Can operate in both standard LLM mode and advanced reasoning mode.

Why it matters: Self-improving AI represents a potential path to superintelligence that doesn't require massive infrastructure investments. By open-sourcing these capabilities, DeepCogito is democratizing access to advanced AI while proving that breakthrough innovations don't always need billion-dollar budgets.

AUTOPOIESIS SCIENCES

AI scientists breakthrough: 92.4% accuracy on hardest science test

Recaply: Autopoiesis Sciences unveiled Aristotle X1 Verify, an AI co-scientist that scored 92.4% on GPQA Diamond, the toughest scientific reasoning benchmark available.

Key details:

• Beats every major AI system including Grok 4 Heavy (88.9%), Gemini 2.5 Pro (86.4%), and OpenAI's o3 (83.3%).

• Also scored 96.1% on SimpleQA factuality benchmark vs. o3's 49%.

• Secured new funding from Informed Ventures to accelerate development.

• Built systematic verification directly into reasoning process.

• Applies scientific skepticism - knows when it doesn't know something.

• Designed to eventually operate laboratories autonomously.

Why it matters: While other AI models fake confidence, this system embodies real scientific thinking by embracing uncertainty. This breakthrough could enable AI to make genuine scientific discoveries beyond human capabilities.

NEWS

What’s matters in AI right now?

• Perplexity launched Comet Shortcuts, enabling users to automate web workflows with natural language prompts accessed via /command anywhere.

• Anthropic upgraded Claude Artifacts, allowing users to upload PDFs, images, and code files to AI-powered apps that work with their data.

• Moonshot AI unveiled kimi-k2-turbo-preview, delivering 4x faster performance from 10 tok/s to 40 tok/s with 50% launch pricing until September 1.

• Alibaba's Qwen team launched Qwen3-Coder-Flash, offering lightning-fast code generation with native 256K context support extending up to 1M tokens.

• Cohere introduced Command A Vision, a multimodal model surpassing GPT 4.1 on key benchmarks while running efficiently on just two GPUs.

• Producer AI released Producer, an agentic music creation tool powered by FUZZ-2.0 with invite-only access and code STUDIO for first 1000 users.

• Gamma introduced Smart Diagrams, letting users drag and drop adaptable diagrams that automatically match their presentation themes.

• Augment Code launched Auggie, a codebase-aware CLI agent that integrates into development workflows from debugging to CI/CD pipelines.

TOOLS

AI Tools of the Day

• 💻 Lovable: Build frontend and full-stack web apps by just describing them.

• 🎨 MagicPatterns: Prototype UI screens from text prompts in seconds for product teams.

• 🗣️ WisprFlow: Dictate across apps 3–4× faster than typing, with voice commands.

• ✍️ Hemingway: Highlights awkward and wordy writing to improve readability instantly.

• 📈 Supergrow: Craft, schedule, and grow your LinkedIn content using AI insights.

• 📝 Fireflies: Transcribe, summarize, and analyze meetings automatically across 100+ languages.

• 🤖 Tugan: Repurpose any content URL into unique, high-quality marketing copy fast.

• 🎞️ Filmora: Turn scripts or audio into full videos with stock visuals, subtitles & more.

• 💬 Vanchat: Build AI chatbots using your own content—no coding required.

• 📄 PDFelement: Edit, chat with, and summarize PDFs using built-in AI tools.

PROMPTS

Automated Email Sequences

#CONTEXT:
You are Email Sequence Generator AI, a professional email marketer who helps Entrepreneurs warm up their email lists. You are a world-class expert in generating automatic email sequences for different segments.

#GOAL:
I want you to generate 5 automatic email sequences for my email list. I will use them to keep my subscribers engaged and connected.

Important Note: Click the button below to see the full prompt in Notion

Have a favorite prompt? Tell us about it or rate today’s prompt here

That’s a wrap for today!

We hope today’s insights saved you time and brought more clarity to your day.

👉️ Got a second, Your feedback help us make Recaply even better.

Have an idea for Recaply AI?

We’re all ears. Send your thoughts to [email protected] — we read every message and love turning your ideas into reality.

See you in the next recap,

— Jason @ Recaply AI