Meet the $5,900 humanoid that can cartwheel

Good morning, AI enthusiasts. China just lobbed a grenade into the humanoid market with a $5,900 full-size robot that can cartwheel downhill, Alibaba open-sourced a giant reasoning model that’s topping open benchmarks, MIT built a Nintendo controller-style tool so factory workers can train bots on the fly, and an Indian studio is about to re-release a 2013 romance, this time with an AI-generated happy ending that the director never approved.

In today’s TLDR AI:

• Unitree’s R1 humanoid drops the floor price 10×, sparking a global robot price war

• Alibaba’s Qwen3-235B open-source model beats GPT-4 on key reasoning tasks

• MIT’s handheld interface lets anyone teach a robot via joystick, kinesthetic or demo mode

• Indian studio uses AI to rewrite a tragic film finale, igniting a creative rights fight

LATEST DEVELOPMENTS

🤖 Unitree’s $5,900 humanoid stuns the market

• TLDR: The new R1 packs 26 joints, flips, sprints and sells for just ¥39,999 (~US $5.9K), cheaper than many e-bikes.

• Demo video shows the 127 cm robot cartwheeling, running downhill and self-righting in seconds, moves usually reserved for $50K+ machines.

• R1 undercuts Unitree’s own G1 (US $13.8K) and Tesla’s projected Optimus (<$20K), setting a new floor for dev-grade humanoids.

• Company filed IPO paperwork days earlier, eyeing funding to mass-produce “tens of thousands” of units in 2026.

• Analysts say a sub-$10K sticker could lure hobby coders and university labs now priced out of humanoid R&D.

• Why it matters: A $6K price tag turns humanoids from Moon-shot toys into Raspberry Pi-level tinkering platforms, accelerating software and competition overnight.

🧠 Alibaba’s Qwen3 tops open benchmarks

• TLDR: The Qwen3-235B-A22B-“Thinking-2507” model scores 79.7 % on Arena-Hard v2, edging every other open LLM and rivaling GPT-4 in reasoning tests.

• Released on Hugging Face + ModelScope under a permissive license with an FP8 “lite” build for single-GPU rigs.

• Outperforms previous open leaders on math, code and multilingual tasks while running inference 1.5× faster than DeepSeek-R1 thanks to hybrid MoE layers.

• First major model to ship with transparent chain-of-thought gating, letting devs audit reasoning steps.

• Chinese devs call it “the open-source answer to GPT-4,” while US labs quietly benchmark against it for internal tool stacks.

• Why it matters: Open-source just closed another gap with frontier pay-walled models, pressuring Western labs and export regulators to rethink their lead.

🛠️ MIT handheld interface lets anyone train a robot

• TLDR: A “versatile demonstration interface” clips onto standard cobot arms so users can teach tasks by joystick, by physically moving the arm, or by simply showing the tool what to do.

• Combines tele-op, kinesthetic and natural demo into one gadget; volunteers preferred “natural” mode for delicate molding tasks.

• Camera + force sensors record motions; robot replays with millimetre accuracy after one pass.

• Tested by manufacturing pros on press-fit and thermo-mold chores; average teaching time fell 60% versus legacy joystick programming.

• Team will open-source CAD files and code after IROS 2025, hoping to bridge the skills gap on factory floors.

• Why it matters: Programming a robot could soon feel like tracing a shape in the air with no coding and no downtime, blurring the line between operator and developer.

🎬 Bollywood film gets AI “happy ending”

• TLDR: Eros Media will re-release 2013 hit Raanjhanaa on Aug 1 with an AI-generated feel-good finale, renaming it Ambikapathy.

• CEO calls it an “exploratory baby step” and hints 3K more titles are under review for similar makeovers.

• Director Aanand L Rai learned via press reports, calling the move “reckless and dystopian” and considering legal action.

• India’s Directors’ Guild says the case may set global precedent for altering finished films without creator consent.

• Posters will carry an “AI-powered ending” badge; Eros won’t say if the disclaimer appears on screen.

• Why it matters: If studios can rewrite endings at will, creative control and even canonical storylines could be one prompt away from disappearing.

COMMUNITY

🎓 JOIN OUR FREE WORKSHOP 

Want to master AI & automations? Ready to turn your AI skills into a new revenue stream? Get started with our FREE workshop.

Learn how you can monetize AI and get certified as an AI specialist during our free web class. Click here to register.