E-Paper
Humanoid Robots, a new Gemini and OpenAI’s house of disorder
The biggest AI developments, decoded. 19 Aug 2026
Cognitive warmup. Chinese robotics company Unitree, perhaps the country’s most exciting one right now (a good place to be, as the IPO evolves to a debut on the Shanghai Stock Exchange), has built a G1 humanoid that can throw kung fu kicks and bounce back from knockdowns. Foundation, a robotics company in the US, is believed to have demonstrated autonomous navigation and human detection capabilities to the Department of Homeland Security, making a case that humanoid robots to patrol the country’s southern border. The US has also banned imports of Chinese humanoid robots including robot dogs, due to fears around data security.
Research firm Smart Analytics Global (SAG) humanoid robot shipments surged 272% year on year, to 19.1K units in 1H 2026. A key here is, industrial and commercial applications now make for more than 70% of industry volume. Also, while I say Unitree Robotics has the most exciting optics, turns out rivals AGIBOT overtook Unitree to lead the market share, with the two companies now holding around 75% of the total shipment share, and AGIBOT has cornered 44% of global shipments.
The research firm expects global humanoid robot shipments to approach 60K units in 2026, nearly tripling year-on-year, with industry revenue reaching approximately US$1.6 billion.
“New segments such as ultra-bionic humanoid robots targeting human interaction and emotional companionship are emerging, although their commercial potential remains to be proven. Regulatory uncertainty around highly human-like robots and AI interactions, together with rising geopolitical risks, could also become increasingly important factors shaping the industry’s next phase of growth,” Linda Sui, Founder and Principal at Smart Analytics Global.
- In July, Samsung sped up plans for the humanoid robotics space by establishing a dedicated Robotics eXperience (RX) division. The company is backing this initiative with a 19 trillion won (~$14 billion USD) investment.
- This month, LG Electronics and Nvidia confirmed collective development of a next-generation humanoid robot. This will use Nvidia’s Jetson Thor chips as well as the Isaac GR00T foundation model (as its brain), and scheduled for debut in Q1 2027.
- In the midst of strong momentum, AGIBOT had released the A3 Ultra humanoid. There is distinct advantage, with the company putting together a full ecosystem (such as the OmniHand and the Quadruped focused on education use-cases as well), including modular research platforms and dexterous robotic hands.
GOOGLE’S NEWEST GEMINI LEAP
Google has released their newest low cost, high speed AI model, Gemini 3.7 Flash, three weeks after Gemini 3.6 Flash. They call it the most “intelligent workhorse model yet for coding and agents.” Of course, they would. But as I’ve often pointed out, Google’s AI conversation isn’t ever without substance of relevance.
Gemini 3.7 Flash marks significant jumps in software engineering tasks—a score of 65.3% (up from 49% for the predecessor model on the DeepSWE v1.1 agent benchmark), and the leap from 34.4% to 43.6% on FrontierCode 1.1 benchmark represents the coding advancements. If you note complex document comprehension, the model marks massive reasoning upgrades for knowledge-dense fields, jumping from 22% to 34% on the GDP.pdf benchmark for dense data extraction.
I’ve often pointed out real-world relevance of Google’s AI developments. That continues with the Gemini 3.7 Flash, and I’ll illustrate some illustrative specifics.
- The Code Arena score of 1588 puts Gemini 3.7 Flash significantly ahead of OpenAI’s GPT-5.6 Terra (1523) and Meta’s Muse Spark 1.2 (1535), as far as web developments capabilities are concerned.
- For the enterprise workflow automation benchmark called AutomationBench, a 30.4% result is much higher than GPT-6.5 Terra (23.6%) and Claude Sonnet 5 (10.7%).
- Harvey LAB-AA benchmark for complex legal workflows returns a score of 90.7%, ahead of Claude Sonnet 5 (90.1%) and GPT-5.6 Terra (85.2%).
- In the LABBench2 Biology real-world research benchmark, Gemini 3.7 Flash returns an 82.1% result, much ahead of GPT-5.6 Terra’s 81.2% and Claude Sonnet 5’s 80.1%.
OPENAI’S EXECUTIVE EXODUS
Fidji Simo, CEO of AGI Deployment at OpenAI. Kate Rouch, first Chief Marketing Officer at OpenAI. Brad Nightcap, COO. Chloé Bakalar, the AI Ethics Lead. That’s a long list of executives who have in recent times, or are preparing, to exit OpenAI. Not a good look for an AI company that tries to portray a success story every now and then.
Add to that list of departures, Denise Dresser, who joined OpenAI as its chief revenue officer in December. Dali Rajic, president and COO of Wiz, will be taking over the CRO role. The bullish positioning continues. “We’re now at an inflection point: the next generation of models will change not just how work gets done but how companies are built and run. Dali will build the revenue operating system needed to scale for this next phase,” the company says.
Most certainly, so many top executives leaving in a short window isn’t at all signalling a position of strength. Since earlier this year, there has been pressure for revenue at OpenAI, and that included cutting side projects. Sora, the AI video generation app was discontinued. OpenAI agreed to a Pentagon contract that rivals Anthropic refused to sign because of concerns about autonomous weapons and domestic mass surveillance. Just a snapshot of the last few months at OpenAI.
