Find out why 100K+ engineers read The Code twice a week

Staying behind on tech trends can be a career killer.

But let’s face it, no one has hours to spare every week trying to stay updated.

That’s why over 100,000 engineers at companies like Google, Meta, and Apple read The Code twice a week.

Here’s why it works:

• No fluff, just signal – Learn the most important tech news delivered in just two short emails.

• Supercharge your skills – Get access to top research papers and resources that give you an edge in the industry.

• See the future first – Discover what’s next before it hits the mainstream, so you can lead, not follow.

Join 100,000+ engineers who read The Code to stay ahead of the curve.

Today:

• OpenAI DevDay 2025: Everything You Need to Know in One Report

• OpenAI Taps AMD for 6GW of GPU Power in Multi-Year Deal

• Google DeepMind Debuts CodeMender to Fix Security Bugs with AI

• AI Startups Took Over Half of All VC Funding in 2025

• Google Tests AI-Generated Snippets—Site Meta Descriptions at Risk

🚀 OpenAI DevDay 2025: Everything You Need to Know

OpenAI’s DevDay 2025 was packed with major updates showing how fast ChatGPT is evolving—from skyrocketing user numbers to powerful new tools for developers and businesses. If you blinked, you might’ve missed the unveiling of agent builders, the booming Codex coding system, and price cuts on nearly everything from video to voice.

🌍 Growth Is Off the Charts

ChatGPT’s user base exploded from 100 million weekly users in 2023 to over 800 million today. Developer numbers doubled to 4 million weekly, and API usage soared to 6 billion tokens per minute, showing just how deeply AI is being integrated into real-world workflows.

🧠 Build Apps Inside ChatGPT

One of the biggest reveals: users can now create fully functional apps directly inside ChatGPT via the new Apps SDK, powered by the Model Context Protocol. Brands like Canva, Coursera, Expedia, Figma, Spotify, and Zillow are already live (outside the EU), with in-chat checkout via the Agentic Commerce Protocol.

The roadmap includes public app submissions, a searchable app directory, and rollout to ChatGPT Business, Enterprise, and Edu customers.

🤖 Agents Are Real Now

OpenAI introduced AgentKit, Agent Builder (a drag-and-drop canvas), and ChatKit (plug-and-play chat components). Developers can connect apps to Dropbox, Google Drive, SharePoint, Teams, and more via the new Connector Registry. Guardrails are also being open-sourced for safety and PII protection.

💻 Codex Graduates

Codex—OpenAI’s coding assistant—is officially out of research preview. With Slack integration, admin tools, and SDKs, it's helping dev teams boost productivity. Since August, usage is up 10×, and GPT‑5‑Codex has handled a stunning 40 trillion tokens in just three weeks.

OpenAI says this has helped their own engineers merge 70% more PRs weekly, with nearly every PR getting automatic review.

💸 API Pricing: Cheaper, Faster, Smarter

OpenAI's 2025 lineup includes new models optimized for speed, accuracy, and cost-efficiency:

• GPT‑5‑Pro: Top-tier accuracy for finance, law, and healthcare — $15 in / $120 out per million tokens.
  

• GPT‑Realtime‑Mini: 70% cheaper text + voice model — as low as $0.60 per million.
  

• GPT‑Audio‑Mini: Built for transcription and TTS — same price as above.
  

• Sora‑2 & Pro: Video generation with sound — $0.10–$0.50/sec depending on quality.
  

• GPT‑Image‑1‑Mini: Vision tasks at 80% less than the large model — just $0.005–$0.015 per image.

OpenAI Taps AMD for 6GW of GPU Power in Multi-Year Deal

OpenAI and AMD struck a multi-year deal to deploy six gigawatts of AMD Instinct GPUs, starting initially with one gigawatt of MI450 chips in late 2026. AMD issued OpenAI a warrant for up to 160 million shares, vesting as deployments and price milestones are met.

KEY POINTS

• Scale & timing: Six gigawatts of GPU capacity; first one-gigawatt MI450 rollout planned for the second half of 2026.

• Aligned incentives: AMD granted OpenAI a warrant for up to 160 million AMD shares; vesting depends on deployment milestones and share-price targets.

• Strategic partnership: AMD becomes a core compute partner across multiple chip generations; AMD expects “tens of billions” in revenue from the deal.

Why it matters

This builds the huge computer power OpenAI needs to train and run stronger AI systems. It also locks in a steady hardware path over several years, helping OpenAI deliver faster, better tools while giving AMD major growth from future AI demand. 

Google DeepMind Debuts CodeMender to Fix Security Bugs with AI

Google DeepMind unveiled CodeMender, an AI agent that finds and fixes software security bugs. It patches flaws quickly and rewrites risky code to prevent future issues. In six months, it contributed 72 fixes to open-source projects, using testing and human review to ensure reliable changes.

KEY POINTS

• What it is: An agent powered by Gemini “Deep Think” that locates root causes, writes patches, and validates them before sending high-quality fixes to humans

• Early results: 72 upstreamed fixes across open-source projects (some ~4.5M lines). It also rewrites code proactively—e.g., adding “-fbounds-safety” so the compiler blocks buffer overflows in libwebp.

• How it stays safe: Multi-step checks (static/dynamic analysis, fuzzing, differential tests) and an LLM “judge” ensure no regressions; all patches still get human review.

Why it matters

Security holes are hard and slow to fix. Automating the hunt and the patch speeds protection for users and reduces zero-day risks. Proactive rewrites can remove whole classes of bugs, easing strain on open-source maintainers while keeping software safer for everyone. 

🧠RESEARCH

Apriel-1.5-15b-Thinker

Apriel-1.5-15B-Thinker is a new open-source AI model that handles both images and text. Despite using less computing power, it rivals larger models like Claude and Gemini in performance. Trained in smart stages without reinforcement learning, it shows that careful training can beat brute force. All tools and data are publicly shared.

StealthAttack: Robust 3D Gaussian Splatting Poisoning via Density-Guided Illusions

Researchers developed StealthAttack, a new way to secretly sabotage 3D scenes created with Gaussian Splatting. By adding fake visual elements in low-density areas, the attack fools certain camera angles while leaving others unchanged. This method is harder to detect and outperforms earlier attack techniques. A full demo is available online.

StockBench: Can LLM Agents Trade Stocks Profitably In Real-world Markets?

StockBench tests if AI agents like GPT-5 or Claude-4 can trade stocks in real-world conditions. Most models failed to beat a simple buy-and-hold strategy, but a few showed promise in managing risk and returns. The benchmark reveals major challenges and opportunities in building AI agents for real financial markets.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

Vmake AI – AI Video Enhancer & Editor - AI video toolset focused on making “talking” and short‑form videos look and sound better—fast.

Upscayl – AI Image Upscaler - AI upscaler that enlarges and enhances low‑resolution images.

Websim AI – AI Website Builder - AI website/app generator and social creation platform.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Venture capitalists invested a record $192.7 billion into AI startups in 2025, with over half of all VC funds going to AI. Big players thrived, while non-AI startups struggled to attract funding.

• Google is testing AI-generated descriptions and summaries for search snippets, replacing website-provided meta descriptions. This shift may affect how pages appear in search results and impact click-through rates and content control for site owners.

• Huawei released an open-source tool called SINQ that shrinks AI models by 60–70% without hurting quality. It lets developers run large models on cheaper hardware, cutting costs and expanding access to powerful AI.

• Google launched a new AI bug bounty program offering up to $30,000 for finding serious flaws like rogue actions in its AI systems. Bugs involving content issues won't qualify and must be reported separately.

• Mattel is partnering with OpenAI to test its Sora 2 AI video model. Designers can now turn sketches into shareable concept videos, speeding up toy development and showcasing AI’s creative potential in product design.

• OpenAI and Allied for Startups released Hacktivate AI, a report with 20 proposals to speed up AI adoption in Europe. It aims to boost skills, cut red tape, and help more sectors use AI effectively.