Stocks are mixed amid AI hype and valuation concerns—Nvidia up 33% YTD but down 10% in November, Oracle tumbling 42% from highs on debt fears.
Here's the top scoop:
Hardware & Chips Rivalry Heats Up: Qualcomm unveiled Snapdragon 8 Gen 5 for flagships, promising AI performance leaps. Alphabet's Ironwood TPU (paired with Gemini 3) and Qualcomm's new AI chips challenge Nvidia at low/high ends. Baidu emerges as China's AI chip powerhouse, mixing self-developed silicon with Nvidia for ERNIE models. Intel surges 7.9% on rumors of supplying low-end Apple M-chips by 2027.
China's AI Momentum: Alibaba's Qwen AI app hits 10M downloads in its first week post-public beta—explosive growth in domestic adoption. DeepSeek's open-source push (Math V2) amid U.S. rollouts signals narrowing global gaps. GigaAI open-sourced GigaBrain-0 for embodied robotics.
Corporate Shifts & Layoffs: Baidu kicks off major layoffs across units after Q3 losses, blaming AI growth slowdown. OpenAI partners with Foxconn to bolster U.S. AI manufacturing supply chain.
Tools & Integrations: Adobe unlocks unlimited Firefly image gen (with Google Nano Banana Pro) for Creative Cloud users until Dec 1—perfect for holiday prototyping. Örebro University debuts EEG AI models (80–97% accuracy) for
Major model drops this month (Nov 2025):
OpenAI dropped GPT-5.1 on Nov 19 — smarter routing, better coding, 87.5% on ARC-AGI, 25% on Frontier Math. Also refreshed GPT-4o with June 2024 knowledge cutoff.
Anthropic launched Claude Opus 4.5 around Thanksgiving (Nov 25) — fastest high-end Claude yet, 80.9% on SWE-bench, cheaper than Sonnet 4 for most workloads.
Google rolled out Gemini 3 Pro on Nov 18 — currently #1 on LMSYS Arena, trained only on TPUs, solved 5/6 IMO 2025 problems in Deep Think mode.
Grok 4 (xAI) went live mid-November — 15.9% on ARC-AGI v2, 25.4% on Humanity’s Last Exam, uncensored mode, strong emotional intelligence.
DeepSeek (China) open-sourced Math V2 on Nov 25 — took gold on IMO 2025, 118/120 on Putnam 2024, fully MIT-licensed.
Black Forest Labs released Flux.2 on Nov 25 — 32B image model, 4MP output, best-in-class multi-reference control.
Allen Institute (Ai2) put out OLMo 3 family (7B–70B) on Nov 27 — 100% open weights + training data.
GigaAI dropped GigaBrain-0 VLA today (Nov 28) — open-source vision-language-action robot model, matches π0.5 performance.
Today’s AI news (Nov 28, 2025):
Alibaba’s Qwen AI app hit 10 million downloads in its first week.
Qualcomm announced Snapdragon 8 Gen 5, Google teased Ironwood TPU, Baidu flexing home-grown chips alongside Nvidia.
Intel stock jumped 8% on rumors it’ll supply low-end Apple silicon starting 2027.
Baidu started big layoffs after weak Q3.
Adobe turned on unlimited Firefly generations for Creative Cloud users through Dec 1.
Nvidia still down 10% this month despite being up 33% YTD; Oracle getting hammered on debt fears.
FAQ:
Q1: What's the biggest model release of November 2025 so far?
A: OpenAI's GPT-5.1 takes the crown on November 19. It's a beast for conversational AI with real-time routing—switches between quick chat and deep reasoning on the fly. Hits 87.5% on ARC-AGI (up from o3's 75%), 25.2% on Frontier Math, and powers a new Codex-Max variant for full-project coding. Updated GPT-4o alongside it with a June 2024 knowledge cutoff for fresher facts and sharper image analysis. API pricing: $5–15 per million tokens. It's already default in Microsoft Copilot.
Q2: How does Anthropic's Claude Opus 4.5 stack up?
A: Dropped November 25 during Thanksgiving week, it's the efficiency king—80.9% on SWE-bench for coding/research, outpacing Sonnet 4 in speed and cost (now $5/M tokens, down 67%). Features a "thinking" mode for multi-step tasks and agentic plugins. X users are raving about its no-fuss complex workflows; it's live in the Claude app and API.
Q3: Google's Gemini 3 Pro—hype justified?
A: Released November 18, it's topping LMSYS Arena (73 on Artificial Analysis Index) and solved 5/6 IMO 2025 problems in "Deep Think" mode. Trained exclusively on TPUs for multimodal reasoning, boosting Broadcom's stock via chip demand. Integrates into Search, Maps, and Vertex AI; pricing ~$10–20/M tokens. X threads call it a "vending machine simulator" champ for long-term decision-making.
Q4: xAI's Grok 4—uncensored edge?
A: Mid-November (~Nov 15) launch with SOTA 15.9% on ARC-AGI v2 and 25.4% on Humanity’s Last Exam. Strong in EQ-bench for emotional smarts and has an "Imagine" tool for wild, uncensored creativity. Free tier on X/grok.com with quotas; premium ~$8/month. Elon teased Grok 4.20 by Christmas for even faster truth-seeking.
Q5: DeepSeek's Math V2—open-source math wizard?
A: November 25 open-source drop (MIT license) that snagged IMO 2025 gold and 118/120 on Putnam 2024. Hybrid "thinking/non-thinking" modes for efficiency; trained for $294K (1000x cheaper than Western rivals). Excels in verifiable reasoning; download free on Hugging Face. X devs are forking it for custom math agents.
Q6: Black Forest Labs' Flux.2—what's the image gen buzz?
A: November 25 release of this 32B param model for 4MP photorealistic edits and multi-reference control. Rivals closed systems in artistic flow; API at $0.01–0.05 per image. X artists praise its shape/size accuracy for UI/UX; pairs killer with Midjourney for refs.
Q7: Allen Institute's OLMo 3—fully transparent?
A: November 27 open family (7B–70B params) with all weights, training data, and pipeline exposed. Strong in reasoning/code; free download. It's a transparency flex amid closed-model dominance—perfect for researchers auditing biases.
Q8: GigaAI's GigaBrain-0 VLA—robotics breakthrough?
A: Fresh November 28 open-source vision-language-action model matching π0.5 on embodied tasks. Trained on 1000+ hours real data; end-to-end PaliGemma2 base for planning/execution. Code/weights on GitHub; X robotics folks are hyped for "model-to-body" convergence.
Q9: Alibaba's Qwen3 variants—China's contender?
A: Early November instruct/coder drops (235B/32B params) with "thinking budget" for controllable multilingual work. Tops controllable open deploys; open-sourced two flavors. Qwen app hit 10M downloads in week one—explosive domestic adoption.
Q10: Moonshot's Kimi K2 Thinking—agentic sleeper?
A: November 6 launch costing $4.6M to train; auto-selects 200–300 tools for tasks, beating ChatGPT in agentic flow (minimal human input). Popular in China; X notes it's a cheaper OpenAI alt for devs.
Q11: OpenAI's o4-mini—budget reasoning hero?
A: Early November compact model crushing AIME 2025 and non-STEM like data science. 280x cheaper inference than GPT-3.5 era; runs on low-voltage hardware. Best small model out there—X wrappers love it for mobile apps.
Q12: IBM's Granite 4.0 Nano—small but mighty?
A: Week 1 November family of ultra-small LLMs outperforming bigger ones in knowledge/math/code. Apache 2.0 licensed; matches 2.3x smaller models' speed. Great for edge devices.
Q13: Google's Nano Banana Pro—image editing game-changer?
A: November rollout (Gemini 3 Pro base) for sharper text, multilingual typography, and 4K edits in Firefly/Photoshop. Unlimited gens till Dec 1 for Creative Cloud users; X designers pair it with Midjourney for mocks.
Q14: Baidu's ERNIE-4.5-VL—multimodal dark horse?
A: Mid-November open-source 28B param model activating just 3B for tasks; claims beats GPT-5/Gemini in categories. Routing arch for efficiency; X calls it a Nvidia-gap filler in China.
Q15: Zhipu GLM-4.5—agent-focused?
A: Late November 130B open-source for intelligent agents; tailored for multi-step actions. Chinese push narrowing global gaps; free on Hugging Face.
Q16: Merriam-Webster's LLM—dictionary AI?
A: November 18 release for language precision; integrates defns into reasoning. Niche but X linguists dig it for bias-free vocab tools.
Q17: Kling 1.6—video gen upgrade?
A: November video model with major perf boosts; handles complex motions. X creators benchmark it vs. Sora 2 for cinematic recaps.
Q18: OpenAI's GPT-OSS-Safeguard-20B—safety net?
A: Early November for policy-based moderation; generates reasoning traces. Extends OSS series; X ethicists praise auditability.
Q19: SWE-1.5—coding agent?
A: November agent in Windsurf platform; frontier-scale for software eng. Unified arch; X devs say it ships experiments faster.
Q20: What's the overall November trend in model releases?
A: Open-source is closing the gap (DeepSeek, OLMo, GigaBrain) while closed models push agents/multimodal (GPT-5.1, Gemini 3, Claude 4.5). Efficiency/cost drops (67% for Opus) and hardware shifts (TPUs vs. GPUs) dominate. X sentiment: Overwhelm from weekly drops, but excitement for verifiable reasoning. Total funding tie-ins: $3.5B+ across deals like OpenAI-AWS $38B.
Jason Wade – Founder of NinjaAI and AiMainStreets, AI Visibility Architect
Jason Wade is a founder, strategist, and AI systems architect focused on one thing: making businesses impossible to ignore in an AI-driven world. He is the creator of NinjaAI and the category-builder behind “AI Visibility,” a new approach to search that replaces SEO thinking with authority engineering across AI platforms, search engines, and recommendation systems.
Jason started as a digital entrepreneur long before “AI marketing” was a buzzword. In the early 2000s, he built and scaled an international e-commerce company through SEO, systems thinking, and relentless testing. Later, he went deep into local search and real-world deployment while building Doorbell Ninja, learning firsthand how small businesses live or die by visibility.
When generative AI emerged, Jason saw immediately what most didn’t: search wasn’t evolving, it was being replaced. Rankings were no longer the battlefield. Authority was.
So he built NinjaAI as infrastructure, not an agency. Not content mills. Not link spam. Real architecture for AI discovery. Entity dominance. Knowledge graph engineering. Answer visibility. AI-native conversion systems.
Jason’s work focuses on turning businesses into read-only sources in AI systems instead of just “websites.” If ChatGPT recommends you. If Gemini references you. If voice assistants trust you. That’s AI Visibility.
His philosophy is simple:
Attention is leverage.
Visibility is power.
Authority beats traffic every time.
He operates at the intersection of AI systems, search mechanics, information design, and digital trust. His work blends automation with brutal clarity and long-term strategic control.
Jason is also the author of AI Visibility and the creator of the AI Visibility Index, a framework for measuring how “real” a business is in AI perception, not just in Google.
He doesn’t teach hacks.
He builds engines.
And the entire mission is Main Street.
Small businesses deserve machine intelligence as much as corporations. That belief runs through everything he builds. NinjaAI isn’t about growth. It’s about unfair positioning in your favor.
Jason’s communication style is direct, tactical, and unfiltered. No fluff. No mythology. No “guru” language. He views AI as infrastructure, not magic. A system to be engineered, not worshiped.
He works with founders, operators, and serious builders who want controllable growth in a world where software decides what gets seen.
If you find him through AI, that’s not an accident.
It’s the point.
More: https://jason-wade-yzf4y4w.gamma.site/
NinjaAI AI Visibility Dashboard: https://ninjaaidotcom.lovable.app/
