- AI Fire
- Posts
- 🇨🇳 China’s New 600B AI Challenger
Sponsored by
China’s AI race got another serious player. StepFun’s Step 5 packs 600B parameters, 1M context, vision, and strong coding chops. And only 27B activate per token!?
What's on FIRE 🔥
IN PARTNERSHIP WITH CLOUDWAYS
Cloudways’ free, two-day online event brings together developers, agencies, and businesses to explore what’s next in AI, modern development, and digital growth.
Featured Day 1 conversation: OpenClaw in the Real World: What Are Developers & Agencies Actually Building?
AI agents look great in demos. But what happens when you put them to work? Join the OpenClaw and Cloudways team to explore real-world use cases across coding, research, client operations, and always-on automation, plus what it takes to run agents reliably in production.
September 22–23 | Free Online Event
AI INSIGHTS
StepFun is a Shanghai AI lab founded in 2023 and backed by investors including Tencent and Qiming Venture Partners. It just released Step 5 Preview, a new 600B-parameter MoE model built for coding, agents, and finance, but only 27B parameters activate per token. Main highlights:
1M-token context window with vision support.
Costs about $1 per 1M input tokens and $2.70 per 1M output tokens.
Runs at roughly 100 output tokens/sec in independent testing.
Scores 67.7 on DeepSWE v1.1, putting it near Kimi K3 and GLM-5.3 in StepFun’s reported coding tests.
Finance is a standout area, with 66.4 on FrontierFinance and 83.3 on DRACO in StepFun’s benchmarks.
StepFun positions the model around the Pareto Frontier: getting the best balance between intelligence and inference cost.
Open weights are planned for October 15, 2026.
StepFun itself is becoming a much bigger player than many people outside China realize. DeepSeek made cheap reasoning a story. Kimi pushed huge context and agents. GLM has been pushing coding/agents. Now StepFun is attacking the intelligence-per-dollar frontier.
PRESENTED BY KODEKLOUD
Build. Break. Fix. Learn.
KodeKloud gives you 1,280+ hands-on labs where you provision Kubernetes clusters, write Terraform configs, build CI/CD pipelines, configure Linux systems, containerize apps with Docker, automate with Ansible, and manage Git workflows.
78+ playgrounds let you experiment freely in sandbox AWS environments, Kubernetes clusters, and CI/CD systems without risk.
190+ courses across DevOps, Cloud, and AI pair theory with hands-on labs at every step.
KodeKloud Engineer and 100 Day Challenges provide real-world job scenarios with automated grading that confirms your solutions work.
Stuck? The 55,000+ member Discord community connects you with peers and instructors ready to help.
Every lab runs in a live environment. You deploy, you troubleshoot, you learn. No videos without context. No simulations. The kind of practice that actually builds confidence because you've done real work, not watched someone else do it.
AI SOURCES FROM AI FIRE
1. Video Included: Lesson 6: Build Your AI Automation in Google Workspace (Zero Code Needed!). Learn how Google Workspace Studio & Gemini can connect Gmail, Drive, Gemini, and Sheets to automate workflows and move work forward without constant manual steps.
2. GPT-6 Astra Turns Messy Recording Into Fully Edited Video. Here’s the Full Workflow. Give GPT-6 Astra a rough recording and it can handle the editing workflow for you: cuts, captions, B-roll, music, animations, and even 3D motion graphics. The full workflow is easy.
3. OpenClaw 2.0 is INSANE: Better, Free & UNLIMITED AI Assistant in Mins (Full Setup Guide). This is a massive, ground-up update. You’ll see the complete setup, what makes OpenClaw 2.0 different, and how quickly you can get your own assistant running.
AGI IS HERE. YOUR PROMPTS NEED TO CATCH UP
AI Fire Webinar #11
GPT-6 Astra is already being talked about as one of those “AGI moments.” Yet many people are still using it exactly like they used ChatGPT two years ago: “Do this.”, “Rewrite that.”, “Follow these 17 rules.”, etc.
And if you prompt a smarter model with old habits, you may NEVER feel how much more capable it actually is…
This week, Alex and Eric will show you how to prompt GPT-6 Astra in a way that actually takes advantage of the intelligence you’re paying for.
→ Goal: better results, less trial and error, and a smarter way to work with ChatGPT.
📅 Wednesday, September 23
⏰ 10:00 PM ET / 7:00 PM PT
🔴 Free to join
P/S: Join us live on Zoom and we’ll send you the webinar notes, prompts, and resources after the session.
TODAY IN AI
AI HIGHLIGHTS
🚀 Jev, the stealth AI model that went viral and had people scrambling for access, is now open to everyone. No waitlist. Start here, or use this beginner setup guide first.
🗺️ Do you know who holds power across the U.S. gov.? Most don't. So Michael Adams built US Gov Graph to map the people & positions behind it. Check it out.
🛡️ Z.ai open-sourced ZCode after privacy concerns over encrypted project snapshots sent to Alibaba Cloud. The feature is now removed, and it says no data was kept.
💳 Meta is opening Muse to third-party Connectors & AI payments through Stripe. For small businesses, that means more ways to serve customers directly inside Muse.
🏛️ Trump announced plans for a federal “AI Force” modeled on Space Force, plus a future AI czar. We still know few details on budget, structure, or where it would sit.
🍽️ Sam Altman, Tim Cook, and Jensen Huang are all heading to Trump’s dinner with Xi Jinping. Dario Amodei, meanwhile, hasn’t been named among the guests yet.
💰 Big AI Fundraising: Manus is reportedly raising $500M at a $4B valuation after returning to independent operations. The AI startup previously topped $100M ARR and is now exploring a Hong Kong IPO.
NEW EMPOWERED AI TOOLS
🍎 Aristotle helps kids solve problems out loud with an AI tutor and shared whiteboard, developed with Stanford for interactive learning.
⚙️ Mycel runs the work your service business sells: clients, deliverables, approvals, invoices. From $299/mo, or self-host free.
🛍️ Minicart gives you an AI team to launch and run your store, handling marketing, shipping, and customer replies just by chatting.
🔎 Answers turns web research into one API call, returning structured JSON with sources to build research agents and data pipelines.
AI BREAKTHROUGH
Qwen just released Qwen-Image-2.1, a new open-weight 7B image model built for both image generation and image editing in one system. Its main features:
Lightweight and fast: 7B architecture with quicker inference, especially for multi-image input.
Native transparency support: Can generate and edit RGBA images, which makes transparent backgrounds, layer-based edits, and text replacement much easier.
Strong editing fidelity: Supports up to 10 reference images and is designed to preserve identity and product details more reliably.
Broad use cases: Works well for panoramas, infographics, virtual try-ons, portraits, products, and typography-heavy images.
It aims to be a more practical middle ground between huge premium models and weaker small open models. The “generation + editing in one model” setup makes it more useful for real workflows.
Native transparent-image support is a big deal for design, e-commerce, and content teams. Qwen-Image-2.1 looks like a strong open alternative for creators who want fast, lower-cost image generation and editing without giving up too much quality.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply