- The Shift
- Posts
- Chinese Labs Break the Cost Curve
Chinese Labs Break the Cost Curve
Plus, 🎥 How to create 5-minute AI avatar videos with Hedra, Gemini 3 Deep Think raises the bar for scientific AI, and more!
In partnership with
Welcome back to The Shift. Let’s get straight to what matters in AI today…
Today we have:
🧠 Open-Source Muscle vs Agentic Efficiency
🎥 How to Create 5-Minute AI Avatar Videos with Hedra
🧠 Gemini 3 Deep Think Raises the Bar for Scientific AI
🔨Tools and Shifts you cannot miss
🧠Open Source isn’t Playing Defense Anymore
Two Chinese labs just dropped serious signals about where the open frontier is heading: bigger systems engineering brains from GLM-5, and brutally efficient, agent-native productivity from MiniMax M2.5
The Shift:
GLM-5
GLM-5 Scales and Competes – GLM-5 pushes open-source reasoning and long-horizon planning into territory rivaling and at times surpassing proprietary leaders like Claude Opus 4.5 and OpenAI, proving open systems can compete seriously on depth, math, coding, and agentic performance.
Native Agent Mode - GLM-5 includes a built-in Agent Mode that turns prompts into structured, ready-to-use outputs across multiple formats. Instead of just answering questions, it executes multi-step tasks and generates production-ready documents and workflows.
MiniMax M2.5
3M2.5 Built For Work - MiniMax M2.5 matches or edges past leading closed models on coding and agentic workflows, while being deeply optimized for real-world productivity across development, research, and office-style tasks.
Agents at Scale - The real disruption is cost and speed. MiniMax makes near-frontier intelligence affordable enough to power persistent, always-on agents, shifting AI from “premium assistant” to operational infrastructure.
This is not just one lab flexing. With GLM-5 pushing long-horizon agent planning and MiniMax M2.5 turning enterprise coding into a near-commodity, Chinese models are attacking both capability and cost at once. The pressure is no longer technical. It is economic, and it is accelerating. You can try GLM-5 and Miimax M2.5 here.
TOGETHER WITH ATTIO
Attio is the AI CRM for modern teams.
Connect your email and calendar, and Attio instantly builds your CRM. Every contact, every company, every conversation, all organized in one place.
Then Ask Attio anything:
Prep for meetings in seconds with full context from across your business
Know what’s happening across your entire pipeline instantly
Spot deals going sideways before they do
No more digging and no more data entry. Just answers.
🎥 How to Create 5-Minute AI Avatar Videos with Hedra
Need a talking head that never messes up a line? Here’s how to use Hedra Avatars:
1️⃣ Create your character – Design your custom avatar with the look and vibe you want.
2️⃣ Upload any audio – Drop in a podcast clip, script narration, or voiceover.
3️⃣ Generate – Hedra produces up to 5-minute uncut videos with perfect lip-sync, natural motion, and full facial expressions.
No filming. No retakes. No editing timelines.
Just plug in audio → get a production-ready talking head. You can try it here.
🧠 Gemini 3 Deep Think Raises the Bar for Scientific AI
Google has upgraded Gemini 3 Deep Think into a frontier reasoning system built with scientists and engineers to handle open-ended research, messy data, and real-world engineering challenges across domains.
The Shift:
🏆 Benchmark Breakthroughs - Deep Think scored 84.6% on ARC-AGI-2, 48.4% on Humanity’s Last Exam, achieved 3,455 Elo on Codeforces, and reached gold-medal level performance at the 2025 International Math Olympiad.
It delivered gold-medal written results in the 2025 Physics and Chemistry Olympiads and scored 50.5% on CMT-Benchmark, demonstrating advanced reasoning across theoretical physics, chemistry, and complex scientific domains.
🛠 Real Applications - Researchers used Deep Think to detect logical flaws in peer-reviewed math papers, model physical systems through code, and transform sketches into precise, 3D-printable design files for real-world engineering tasks.
Google introduced Aletheia, an autonomous math agent solving open problems and verifying proofs, while releasing Deep Think to Google AI Ultra subscribers and offering API early access to researchers and enterprises.
Deep Think signals Google’s push toward research-grade AI that competes not only on chat fluency but on measurable scientific rigor. The combination of benchmark dominance, autonomous math exploration, and real-world engineering integration marks a shift toward AI as a serious research collaborator.
TOGETHER WITH WISPR FLOW
Better input, better output
Voice-first prompts capture details you forget to type. Wispr Flow turns speech into clean prompts you can paste into your AI tools for faster, more useful results. Try Wispr Flow for AI.
🔨AI Tools for the Shift
🤝 EVY – Your AI co-creator inside any app, helping you think, write, and build without constantly switching tools.
🧭 Actionbook – Give your AI agent up-to-date action manuals and live DOM structure so it can operate any website instantly without guessing.
📣 Corippl – Cross-promote newsletters and content with aligned creators in your niche, unlocking authentic audience growth through mutual promotion.
🏛️ Levr – Access distilled boardroom playbooks once reserved for unicorns and top VCs, turning “what should we do?” into decisive execution for your business.
🥗 fuelOS – Track macros instantly with AI-powered food recognition built for mindful eaters who want zero friction in their daily nutrition.
🚀Quick Shifts
🔊 Google is adding Gemini-powered Audio Summaries to Docs, letting users listen to condensed overviews of long documents directly from the Tools menu, now rolling out to select business and AI subscribers.
👓 Meta sold over 7 million smart glasses in 2025, tripling total sales from 2023 and 2024 combined, putting EssilorLuxottica’s 10 million annual target by 2027 well within reach despite likely short-term price pressure.
🤖 Anthropic is expanding Claude’s free tier with file editing, Skills, and third-party integrations, aiming to attract users as OpenAI rolls out ads in ChatGPT.
💼 IBM plans to triple U.S. entry-level hiring in 2026, redesigning roles to focus less on automatable coding tasks and more on customer engagement, signaling a bet on human talent despite AI-driven job disruption forecasts.
💰 Anthropic closed a massive $30B Series G round, boosting its valuation to $380B as it races against OpenAI to dominate enterprise AI demand with Claude and enterprise-grade models.
🧩 How to Map Barriers to Entry and Remove Buying Friction Using One Prompt
Turn hidden hesitation into clear, fixable obstacles that prevent first-time buyers from purchasing.
Paste the prompt: Drop this into ChatGPT, then fill in your product and target audience.
Prompt to paste:
Identify the main barriers preventing first-time buyers from purchasing [Insert product]. For each barrier, explain why it exists and provide a clear solution to reduce or remove it. Keep the output practical, conversion-focused, and prioritized by impact on purchase decisions.
That’s all for today’s edition see you tomorrow as we track down and get you all that matters in the daily AI Shift!
If you loved this edition let us know how much:
How good and useful was today's edition |
Forward it to your pal to give them a daily dose of the shift so they can 👇
Reply