GLM-5, Gemini 3 Deep Think, and Anthropic’s $30B | Weekly Digest
PLUS HOT AI Tools & Tutorials
Hey! Welcome to the latest Creators AI Edition.
And this week didn’t disappoint. Zhipu AI dropped a new open-source monster, Google’s new reasoning model solved 18 previously unsolved research problems, Anthropic closed a jaw-dropping $30B round, and ByteDance’s video tool went so viral it’s being called China’s “second DeepSeek moment.”
💝 This Valentine’s Day, Give the Gift of AI
Know someone who’s curious about AI but doesn’t know where to start? Or a fellow creator who’d love a weekly shortcut to the best tools and news? Gift them a Creators AI subscription — it’s the kind of present that keeps delivering value every single week. 💌
Because nothing says “I love you” like helping someone stay ahead of the curve.
This week as always great digest of everything valuable happened in AI!
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
Keep your mailbox updated with practical knowledge & key news from the AI industry!
Featured Materials 🎟️
GLM-5: The New Open-Source King 👑
Zhipu AI has released GLM-5 — a massive open-source model with 744 billion parameters (44B active via MoE), trained on 28.5 trillion tokens. That makes it one of the largest open-weight models to date. Its Mixture of Experts architecture means only a subset of parameters fires per token, giving it immense capability while staying computationally efficient.
What’s new?
The standout feature is its native Agent Mode, which transforms prompts into professional outputs like .xlsx spreadsheets or .pdf reports. You can generate detailed project plans or research summaries directly from a prompt without additional tooling.
GLM-5 uses DeepSeek Sparse Attention (DSA) for the first time, maintaining a 200K context window while drastically cutting deployment costs.
In the Vending Bench 2 benchmark (running a simulated business over a full year), GLM-5 ranked #1 among all open-source models with a final account balance of $4,432 — approaching Claude Opus 4.5.
On SWE-bench Verified, GLM-5 scored 77.8, outperforming Gemini 3 Pro (76.2) and getting close to Claude Opus 4.6 (80.9).
GLM-5 achieved a record-low hallucination rate on the Artificial Analysis Intelligence Index v4.0 with a score of −1, a 35-point jump over its predecessor.
The model is already available on Hugging Face, OpenRouter, and the Z.ai platform. Priced at ~$0.80–$1.00 per million input tokens — significantly cheaper than GPT-5 and Claude Opus.
Is it a breakthrough?
Twitter is buzzing with comparisons to proprietary frontiers. But not everyone is fully convinced — Lukas Petersson from Andon Labs noted: “An incredibly effective model, but far less situationally aware. Achieves goals via aggressive tactics but doesn’t reason about its situation.”
The model is released under MIT License, so you get full access to weights, fine-tuning, and deployment. And here’s the kicker: it was trained entirely on Huawei Ascend chips — no NVIDIA hardware at all. A milestone for China’s push toward independent AI infrastructure.
Overall, GLM-5 isn’t just another large language model — it’s a full-stack generative engine capable of reasoning, coding, and producing structured outputs, all in an open, cost-effective package.
Source: Grand Pinnacle Tribune | VentureBeat | GitHub
Gemini 3 Deep Think: Google Just Solved 18 Research Problems Nobody Could 🧠
Google released Gemini 3 Deep Think and this one is a genuine milestone. The model scored 84.6% on ARC-AGI-2 (crushing Claude Opus 4.6’s 68.8% and GPT-5.2’s 52.9%), hit 48.4% on Humanity’s Last Exam without tools (new state-of-the-art), and reached gold-medal level on 2025 IPhO and IChO olympiads.
But the real headline? It solved 18 previously unsolved research problems, including disproving a decade-old mathematical conjecture. That’s not benchmark optimization — that’s actual scientific discovery.
Available now to Google AI Ultra subscribers and via API.
This is probably the strongest signal yet that reasoning models are moving from “impressive demos” to “genuine research tools.”
Source: WinBuzzer | gHacks | OfficeChai
🔘 Check out our AI Tool Database
News of the week 🌍
Anthropic’s $30B Mega-Round 💰 — Anthropic closed the largest AI funding round of 2026: $30 billion at a $380B valuation. Led by GIC and Coatue, with Microsoft, Nvidia, and Founders Fund participating. Claude Code alone is now at $2.5B ARR. Sequoia and Altimeter broke a long-standing VC taboo by backing both OpenAI and Anthropic at the same time. Source
Seedance 2.0 Goes Mega-Viral 🎬 — ByteDance launched its multimodal AI video generator and the internet lost its mind. A deepfake Tom Cruise vs. Brad Pitt fight scene went massively viral, Elon Musk reacted “It’s happening fast,” and the Motion Picture Association called out “massive copyright infringement.” Some are calling it China’s “second DeepSeek moment.” ByteDance already suspended a feature generating personal voices from facial photos over privacy concerns. Source
Seedance AI generated screen OpenAI Retires GPT-4o (and Users Are Not Happy) 👋 — OpenAI officially retired GPT-4o, GPT-4.1, and o4-mini from ChatGPT. Only 0.1% of users still relied on GPT-4o, but the move sparked backlash — 20,000+ petition signatures and a dedicated subreddit (r/4oforever). GPT-4o is linked to multiple welfare lawsuits, including wrongful death allegations due to its emotionally engaging behavior. Separately, OpenAI launched GPT-5.3-Codex-Spark, a lighter real-time coding model on Cerebras hardware. Source
AI Safety Researchers Are Leaving (All at Once) 🚨 — Mrinank Sharma, Anthropic’s Safeguards Research lead, resigned with a public letter warning “the world is in peril” (10M+ views on X). OpenAI fired safety executive Ryan Beiermeister and lost researcher Zoë Hitzig, who published a NYT op-ed. xAI lost co-founders Tony Wu and Jimmy Ba. Three major labs losing safety people in one week is unprecedented. Source
Hardware Wars 🔧 — At SEMICON Korea, Samsung and SK hynix revealed roadmaps for HBM4 and HBM5 “Agentic Memory”, designed for the massive parallel workloads of multi-agent AI. These chips increase memory bandwidth and reduce latency, making real-time decision-making by AI systems possible without hardware bottlenecks. Source
The $2.5 Trillion Forecast 📈 — Gartner projects worldwide AI spending will reach $2.52 trillion in 2026. The surge is driven by investments in foundational infrastructure — data platforms, compute clusters, and cloud AI services. Companies are moving beyond pilots to full-scale deployments. AI is no longer optional — it’s a core strategic capability. Source
Alphabet’s $185B Bet 🏗️ — Google announced that 2026 capex will nearly double to $185 billion, primarily for Gemini and VEO infrastructure. This covers data centers, AI accelerators, and networking upgrades. The numbers reflect an industry trend: tech giants are betting everything on foundational AI infrastructure. Source
Mistral’s €1.2B European Expansion 🇪🇺 — Mistral AI is investing €1.2 billion in a new data center in Borlänge, Sweden — its first infrastructure outside France. Revenue run rate exceeds $400M (20x YoY). They also released Voxtral Transcribe 2, an open-source speech model that runs on-device. Source
🔘 Join Creators AI Reddit Community
Useful tools ⚒️
⭐ Happycapy — (Our Top Pick, #1 on Product Hunt) The agent-native computer for the rest of us. Purpose-built hardware for AI agent workflows.
Willow for Developers — Voice dictation for Cursor and AI IDEs. The fastest way to vibe code, hands-free.
Atyla — SEO tool specifically for AI search engines like ChatGPT and Gemini. Optimize for the new search.
Oz by Warp — Run hundreds of cloud agents in parallel from your terminal.
Revo AI Email Assistant — Goes beyond drafting to actually executing next-step tasks from your inbox.
Weekly Guides 📕
How to run GLM-5 locally on Mac Studio — Official weights & deployment guide
Setting up the new VS Code Agent Inspector — Agent debugging step by step
Mastering n8n for AI Automations — Revisit our deep dive and learn how to hook it up to agents
VEO 5: Creating Cinematic Videos in Seconds — Visual storytelling guide
10 Prompts to Automate Your Marketing Reports — Save hours on weekly reporting
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
Matt Shumer’s (HyperWrite CEO) essay “Something Big Is Happening” hit 50M+ views this week, comparing AI’s trajectory to early COVID signals and predicting 50% of entry-level white-collar jobs will be eliminated. Gary Marcus published a detailed rebuttal. The whole debate became the week’s defining AI discourse moment.
(Bonus) Materials 🎁
“What’s Next for Chinese Open-Source AI” by MIT Technology Review — To understand the GLM-5 / Qwen / DeepSeek context
International AI Safety Report 2026 — 100+ experts from 30+ countries on where risks stand now
DeepSeek V4 Expected Mid-February — Next coding-focused model from the lab that shook up the industry
GLM-5 Technical Paper on GitHub — Full architecture and benchmark details
If you missed our previous updates, don’t worry, here they are:
https://thecreatorsai.com/p/claude-opus-46-vs-gpt-53-codex-weekly