Code ❤️ Claude

Along with- Phi-4-multimodal and Phi-4-mini

Hi there 👋

Claude just dropped their best model yet - a new ‘reasoning model’ along with Claude Code. What’s really catching attention isn’t just the model itself, but how developers are loving the front-end and the applications it’s generating. 

Have you tried it yet? Let us know your experience in the comments.

Following up on last week, I tested multiple deep research tools for market research- Grok, ChatGPT, and Perplexity. ChatGPT delivered the best results by far- detailed, well-structured, and actually useful. Grok and Perplexity, on the other hand, felt more surface-level.

If you're curious about how these tools stack up, our team has put together a detailed breakdown- check out The Algorithm section for the links!

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Tools - Interesting finds and launches

  • The Algorithm - Resources for learning

  • The Output - Our reflection

Table of Contents

Anthropic has been rolled out some of the most innovative AI features - Claude Artifacts was just the start. Now, they’re keeping up the momentum with Claude 3.7 Sonnet, their most advanced model yet, built with a new Hybrid Reasoning capability.

What is Hybrid Reasoning?

Most AI models either prioritize speed or depth- but Claude 3.7 adapts dynamically. For quick, straightforward questions, it responds instantly. But when it comes to complex problem-solving- like strategic business planning or multi-step coding workflows- it shifts into structured, step-by-step analysis. This makes it particularly powerful for research, finance, and automation.

What’s New in Claude 3.7 Sonnet?

  • Extended Thinking Mode – Users can adjust how much reasoning time the AI spends, balancing speed and depth.

  • Smarter Coding & Debugging – Excels at agentic coding, supporting everything from planning to full development workflows.

  • Better Instruction Following – Achieves 93.2% accuracy in tool use, outperforming competitors in real-world applications.

  • Multimodal Reasoning – Processes structured data, charts, and complex documents for better decision-making.

  • Enterprise AI Integration – Handles both quick responses and deep analysis, reducing the need for multiple AI models in a business setup.

Claude Code: AI That Works Right in Your Terminal

Alongside Claude 3.7, Anthropic is also launching Claude Code- a command-line AI agent that automates software engineering tasks directly in the terminal.

Key Features of Claude Code:

  • Terminal-Based Coding – Edits files, writes tests, and creates commits straight from the command line.

  • Agentic Workflow Execution – Automates tasks while keeping human reviewers in control.

  • Hands-on Code Review – AI suggests code changes, but developers approve them before implementation.

  • Seamless API Integration – Works across GitHub, cloud platforms, and local repositories.

With Claude 3.7 Sonnet and Claude Code, Anthropic is taking AI reasoning and automation to a whole new level, bridging the gap between speed, accuracy, and real-world usability. (source)

Microsoft has introduced Phi-4-Multimodal and Phi-4-Mini, the latest additions to its small language model (SLM) family, designed to enhance AI efficiency and accessibility.

Key Features

  • Phi-4-Multimodal - A 5.6B parameter model that integrates speech, vision, and text processing for context-aware AI applications.

  • Phi-4-Mini - A compact 3.8B model, optimized for text-based tasks such as reasoning, math, coding, and function calling.

  • Optimized for Edge AI - Designed for low-latency, compute-constrained environments, enabling on-device processing.

Why It Matters

Phi-4-Multimodal achieves a notable performance in ASR, vision, and speech translation, surpassing models like WhisperV3 and SeamlessM4T-V2-Large. Phi-4-Mini, despite its small size, matches or exceeds larger models in instruction-following, long-context reasoning, and integration with external tools.

Both models are available on Azure AI Foundry, Hugging Face, and the NVIDIA API Catalog, offering developers flexibility in fine-tuning and deployment. (source)

Amazon Introduces Alexa+: A Next-Gen AI Assistant

Finally - Amazon has upgraded Alexa!

Amazon has announced Alexa+, a generative AI-powered assistant designed to offer more natural conversations, advanced task automation, and deep personalization.

Key Features

  • Conversational AI - Understands casual speech, incomplete thoughts, and complex ideas.

  • Task Automation - Manages reservations, shopping, smart home devices, and more.

  • Personalization - Remembers user preferences, dietary needs, and shopping habits.

  • Cross-Device Integration - Works seamlessly across Echo devices, mobile, browser, and car systems.

  • Privacy & Security - Built on AWS with user-controlled settings.

Why It Matters

Alexa+ expands voice assistant capabilities by integrating agentic AI for task completion, making interactions more intuitive and action-oriented. (source)

Scribe is the first Speech-to-Text (ASR) model, delivering transcription accuracy across 99 languages. It features word-level timestamps, speaker diarization, and audio-event tagging, making it ideal for meetings, subtitles, and real-time transcription.

Why It Stands Out

  • Best-in-Class Accuracy - Outperforms Gemini 2.0 Flash, Whisper Large V3, and Deepgram Nova-3 in benchmarks.

  • Supports Underserved Languages - Reduces errors in Serbian, Cantonese, and Malayalam, where other models struggle.

  • Seamless API Integration - Provides structured JSON transcripts with speaker labels & non-speech event markers. (source)

DeepSeek Fast-Tracks R2 AI Model as China Ramps Up AI Push

As per Reuters, Chinese AI startup DeepSeek is accelerating the launch of its R2 model, originally planned for May, following the success of R1, which triggered a $1 trillion market shift last month. The upcoming model aims to enhance coding capabilities and expand reasoning beyond English, signaling China’s increasing focus on AI leadership.

Built with less-powerful Nvidia chips, R1 rivaled models from U.S. tech giants, raising concerns among Western AI leaders and policymakers. The U.S. government sees AI leadership as a national priority, while Chinese firms have already begun integrating DeepSeek’s models into their products.

DeepSeek, led by hedge fund billionaire Liang Wenfeng, operates more like a research lab than a traditional tech company, with a decentralized structure uncommon in China’s corporate landscape. Industry experts believe R2’s launch could further challenge the dominance of U.S. AI firms, intensifying global AI competition. (source)

Alibaba has made its Wan 2.1 AI model- capable of image and video generation- publicly accessible. It is the first video generation model supporting text effects in both Chinese and English. The four model variants (T2V-1.3B, T2V-14B, I2V-14B-720P, and I2V-14B-480P) are now available on Alibaba Cloud’s ModelScope and Hugging Face for research and commercial use.

This move strengthens Alibaba’s AI presence as it competes in the rapidly growing market. The company also introduced QwQ-Max, an upcoming reasoning model, and announced a $53 billion AI and cloud investment over the next three years.

With Alibaba stock up 58% this year, analysts remain bullish. Citi’s Alicia Yap recently raised her price target from $138 to $170, citing Alibaba’s AI advancements and growing cloud infrastructure. (source)

Deep Research is now accessible to ChatGPT Plus, Team, Edu, and Enterprise users, enabling AI-powered research that analyzes hundreds of sources to generate detailed reports in minutes.

What’s New?

  • Embedded images with citations for richer insights.

  • Improved file understanding for better reference integration.

Users can now share deep research links or request AI-generated reports directly. (source)

Google has introduced a free global version of Gemini Code Assist, offering AI-powered coding help with up to 180K code completions per month, 128K token context window, and support for all public programming languages.

Key Features

  • AI-Assisted Coding - Generates, reviews, and refines code in Visual Studio Code, JetBrains IDEs, Firebase, and Android Studio.

  • Advanced Code Review - Gemini Code Assist for GitHub provides AI-powered pull request feedback for public and private repositories.

  • Natural Language Prompts - Developers can generate and debug code using simple text commands.

  • Custom Style Guides - Teams can tailor code reviews with personalized rules.

Unlike other free coding assistants with strict limits, Gemini Code Assist offers 90x more completions and removes chat restrictions, ensuring students, freelancers, and startups can build without interruptions. (source)

Google has unveiled AI Co-Scientist, a multi-agent AI system powered by Gemini 2.0, designed to assist researchers in hypothesis generation, research planning, and biomedical discovery.

Key Features

  • AI-Driven Research Hypothesis Generation - Creates novel research proposals based on scientific literature.

  • Self-Improving System - Uses iterative AI reasoning, ranking, and evaluation to refine hypotheses.

  • Real-World Validation - Successfully identified new drug repurposing candidates, treatment targets for liver fibrosis, and antimicrobial resistance mechanisms.

Why It Matters

The AI Co-Scientist aims to speed up scientific breakthroughs by automating complex reasoning, bridging interdisciplinary gaps, and assisting researchers in high-impact fields. (source)

Microsoft has unveiled Muse, its first generative AI model for gameplay ideation, capable of generating both game visuals and controller actions. Developed in collaboration with Xbox Game Studios’ Ninja Theory, Muse is built on a World and Human Action Model (WHAM) framework to assist game developers in creating and exploring new gameplay sequences.

Key Features

  • AI-Powered Game Simulation - Generates dynamic game visuals and player actions.

  • Open-Source WHAM Demonstrator - Interactive tool for prompting and testing AI-generated gameplay.

  • Consistency, Diversity & Persistency - Ensures realistic movement, varied gameplay, and user-modified elements persist in sequences.

Muse's open-source model weights and sample data are now available on Azure AI Foundry, allowing researchers and developers to explore new possibilities in AI-driven game design. (source)

Wispr Flow

This tool helps you write 3x faster using just your voice, working seamlessly across all applications. It supports AI-powered commands, auto-edits, and multilingual transcription, making writing more efficient and accurate.

How to Access:

Step 1: Download and install Wispr Flow on your computer.
Step 2: Grant permissions for microphone access and enable Flow in your preferred applications.
Step 3: Start speaking- Flow transcribes your words in real-time, adapting to your writing style and the app you’re using.
Step 4: Use AI commands to edit, format, or rewrite text by selecting it and asking Flow to refine it.

Flow helps eliminate typos, improves accuracy, and breaks writer’s block, letting your ideas flow naturally.

  • Last week, we launched more free AI & ML courses to help you gain hands-on experience with cutting-edge AI tools and frameworks.- 

  • The Deep Research Problem explores the challenges of OpenAI’s Deep Research tool, highlighting how it impresses but also fails in crucial ways. The essay examines its struggle with precise data retrieval, questioning whether AI can ever fully replace human expertise in research and analysis.

  • If you've ever wondered how deep research tools like Grok, ChatGPT, and Perplexity compare for market research, our team has tested them all. We've put together a detailed breakdown to show how they stack up- exploring accuracy, depth, and efficiency.

I’ve been playing around with some projects on Replit, Lovable, Bolt- just experimenting and building. I’ll share some of what I’ve been working on soon.

Also, there’s a lot of buzz about OpenAI possibly dropping ChatGPT 4.5 this week. What’s your take? I’m definitely excited to see what’s coming!

Until next time!

Reply

or to participate.