QWEN3, GPT-5, Gold in Math | Weekly Edition

PLUS HOT AI Tools & Tutorials

Jul 25, 2025

Hey there! Welcome to another Creators’ AI Edition.

This week, Alibaba’s Qwen3 stuns the open-source scene, OpenAI and DeepMind battle for math supremacy, and GPT-5 will be launched in August, Google launches its fastest Gemini yet. GitHub Spark promises instant full-stack AI apps, AI copilots cut clinic errors in Kenya, and ancient history gets a boost from DeepMind’s Aeneas model. Plus: Trump’s latest AI strategy targets chip exports, and a fresh lineup of must-try AI tools, guides, and bonus reads is here to supercharge your week.

Ready to catch the latest AI breakthroughs? Let’s dive in!

This Creators’ AI Edition:

Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁

Featured Materials 🎟️

QWEN3 Update(Beats Kimi K2)

Alibaba’s Qwen team has just set a new bar for open-source AI with their updated, non-thinking Qwen3 model, which now outperforms Kimi K2 and even challenges leading closed models like Anthropic’s Claude Opus 4. Responding to community feedback, Alibaba split its hybrid thinking design and trained instruct and reasoning models separately. The result is a model that activates 22B of 235B parameters and offers a huge 256K context window, leading to major gains across key benchmarks. Qwen3 not only beats Moonshot AI’s Kimi K2, but also holds its own against top frontier models like Claude Opus 4 and GPT-4o-0327. Fully open-source and available as the free default on Qwen Chat, Alibaba’s ChatGPT alternative, Qwen3 highlights how Chinese AI teams are pushing global innovation forward—not just with tech advances, but with a strategic focus on openness and influence, even in the face of Western chip restrictions (more about restrictions in block News of the week).

As usual, Bouncing ball test :)

OpenAI and DeepMind at Math Performance

Google DeepMind and OpenAI have both announced that their latest models have reached gold-medal level performance on the 2025 International Mathematical Olympiad, marking a huge leap in AI’s mathematical reasoning. DeepMind’s advanced Gemini with Deep Think was officially tested by IMO under real competition conditions, solving five out of six problems in algebra, combinatorics, geometry, and number theory, and scoring 35 out of 42 points—the gold-medal threshold—with answers graded and certified by IMO coordinators. OpenAI, meanwhile, ran its evaluation using the same problem statements and rules, also solving five problems and earning a gold-level score, with former IMO medalists serving as graders. While OpenAI’s claim has sparked debate since it was not officially certified by IMO, both results show how quickly AI is closing the gap with top human mathematical talent.

The conversation now shifts from whether AI can solve Olympiad-level problems to when it will start solving math challenges that no human has ever cracked.

News of the week 🌍

GPT-5 Soon

OpenAI is reportedly gearing up to launch its next-generation GPT-5 in August, according to a report from The Verge—with CEO Sam Altman teasing the model this week in a conversation with comedian Theo Von. GPT-5 will merge language skills with o3-style reasoning into a single system, so users won’t have to pick between models for different tasks. Altman described testing the new model as a “here it is moment,” saying GPT-5 instantly answered questions that made him feel “useless relative to the AI.” He also confirmed GPT-5 will arrive “soon,” but clarified it won’t include the specific capabilities used to win the recent gold medal at the IMO competition. OpenAI is also planning to release its first open-weight model since 2019 by the end of July, after delaying its initial launch for extra safety tests.

Both GPT-5 and the open-weight model have sparked months of hype and high expectations, and with Altman talking about potentially “mind-blowing” shifts, August could bring the next major leap in AI.

Gemini 2.5 Flash-Lite is Now Available

Gemini 2.5 Flash-Lite is now officially stable—Google’s fastest, most affordable 2.5 model at just $0.10 per million input tokens. With a 1M-token context window, optional reasoning mode, and ultra-low latency, it’s built for real-time tasks like translation, classification, and video processing. Flash-Lite is already powering tools from Satlyt, HeyGen, DocsHound, and Evertune.

Try it now in Google AI Studio or Vertex AI.

GitHub Spark in public preview

GitHub just dropped Spark, a new tool that turns natural language into full-stack AI apps, frontend, backend, deployment, and all in minutes. Powered by Claude Sonnet 4, Spark handles hosting, data, GitHub auth, and LLMs (OpenAI, Meta, xAI, etc.) with no config or API keys needed. One-click deploys, Copilot integration, and full repo generation make it a game-changer for builders. Available now in public preview for Copilot Pro+ users.

AI Copilots Reduce Errors in Nairobi Clinics

OpenAI has teamed up with Penda Health to study how AI copilots can improve care in Nairobi clinics, finding that clinicians using the AI system made significantly fewer diagnostic and treatment errors compared to those working without it. The AI Consult tool operates in real time, flagging potential issues but letting doctors stay in charge of care. In a study covering nearly 40,000 patient visits, the AI-supported clinicians saw a 16% drop in diagnostic mistakes and 13% fewer treatment errors. All the clinicians surveyed reported improvements in quality, with most calling the AI a substantial “safety net” and educational aid. The research points to three key factors for success: using advanced models like GPT-4o, seamless integration that doesn’t disrupt workflow, and hands-on, personalized training.

This project highlights not only AI’s promise for better healthcare in underserved areas, but also practical steps clinics everywhere can follow to make sure new tools improve care without overcomplicating staff routines.

Missed how AI is helping people in medicine? No worries, here’s one of our posts:

ChatGPT Will Be Your Doctor?

Creators' AI

May 19

Read full story

Trump’s AI Action Plan Targets Chip Exports to China

The Trump administration’s new AI Action Plan calls for tougher export controls on AI chips to stop adversaries, especially China, from benefiting from U.S. innovation. But the plan is vague on how it will enforce or coordinate these controls. While the strategy hints at future executive orders and global alignment efforts, recent contradictions, like greenlighting chip sales to China after restricting them, signal the road ahead is still unclear.

AI Model Aeneas by DeepMind

Aeneas is the first AI model built to contextualize ancient Latin inscriptions, helping historians restore, date, and locate fragmented texts with unprecedented accuracy. Developed with top universities, it uses text and image inputs to find parallels and generate historical insights. By comparing fragments against a massive database, Aeneas turns manual scholarly work into seconds-long tasks. Its open-source release could reshape how we study ancient history—bringing scale, speed, and statistical grounding to epigraphy.

Useful tools ⚒️

Levio by Jupitrr AI - Your AI video editing agent

Submagic API - Generate amazing captions for your shorts with AI

Ash - The first AI designed for therapy (announced $93M total funding)

Opal - Create and share AI mini-apps using natural language

Microtica AI Incident Investigator - An AI Agent that tells you why your systems break

Microtica’s Incident Investigator is an AI Agent that tells you why your systems break. It analyzes logs, deploys, and configs to surface the root cause — fast. No more dashboard hunting. Just context, clarity, and confidence in your incident response.

Share this post with friends, especially those interested in AI!