AI Kung Fu: Better, Faster, Cheaper
June 11, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
In the last few weeks, I’ve noticed some product announcements about speeding up smart software.
On item was “This PCIe AI Accelerator Card Can Run 700B LLMs Locally With 384 GB Memory at Just 240W, Less Than Half The Power of RTX PRO 6000 Blackwell.” Is the “card” real? I have no idea, but the write up makes it clear that Taiwan’s Skymizer is thinking about add-ons that might change the math for some AI use cases… it it works.
I also spotted “New Device Could Make Processors Run 1,000 Times Faster without Additional Waste Heat — Scientists Say It Could Reduce Data Center Energy Demands.” Less heat means lower costs. Good news if one is arm wrestling chip centric AI systems.
Thanks, Midjourney. Four tries to get the word “cheaper” spelled correctly. Not even good enough.
Two examples, one from Japan and one from Taiwan. I assume other silicon whiz kids are beavering away in other shops around the world. I am not sure when, but I think there will be some useful hardware and software that can benefit AI applications and reduce costs… sometime.
VentureBeat, one of my favorite sources for content marketing juiced novelty, has offered another example of this push to make AI better, faster, and in some ways cheaper. “MiniMax-M3 Debuts, Eclipsing GPT-5.5 and Gemini 3.1 Pro on Key Benchmark Performance for Just 5-10% of the Cost” says:
The company’s leadership also announced plans to deliver the model under an open source license including “open weights,” allowing for full enterprise downloading and customizability free-of-charge, coming sometime in the next 10 days. For now, it is available via the MiniMax API at a special discounted price of $0.3 per 1 million input tokens and $1.20 per million output tokens (on fresh cache) for the next week — beating proprietary U.S. giants like Google, OpenAI and Anthropic handily on cost, while also eclipsing the performance of the latest models from the former two on selected benchmarks.
As US vendors push up fees, this Chinese outfits uses the “does more with less” angle and then slams the front part of this digital iron maiden. Oooof. The write up includes some razzle dazzle about how the costs are kept low. No problem, particularly if a company is Chinese and just maybe linked to China’s funding of mission-critical projects.
The write up also trots out some benchmarks that “show” how fast the software is. But the whipped cream on the marketing cup cake is that the MiniMax M3 includes “agentic capabilities.” Here’s an explanation from the VentureBeat write up, and I will leave it to you to deconstruct this chunk of prose:
The system relies on a “Producer + Verifier” adversarial harness loop. As one agent instance generates code, a secondary verifier instance aggressively tests and reflects upon execution outputs, allowing the network to self-correct and operate autonomously for days without human oversight. Because of its native visual grounding, MiniMax Code supports direct computer use. A developer can issue a cross-application voice prompt via their phone to have the model open a localized enterprise ERP client and batch-populate data tables directly from an open Excel spreadsheet. For custom setups, developers can pipeline M3 directly into existing workflows using an API key (
sk-cp) compatible with common alternative IDE environments like Claude Code, Cursor, Roo Code, and Cline. The API introduces a toggleable “thinking mode”.
My view of this announcement is that the Chinese firm wants to make clear that Chinese firms innovating in China are doing what China does: Better engineering.
Several observations are warranted:
- The AI battles are changing. Instead of the slow, ponderous tank-like systems, China is creating more agile solutions. This is similar to Russia relying on tanks and Ukraine recycling hobbyist drones to blow up pipelines. It’s a new era of warfare in AI too.
- The combination of the baby steps with speed ups and modern AI systems from China means that big, quickly out dated data centers might be the anchor than sinks the Silicon Valley super athletes in the long distance swim. Outmoded or not, those big expensive data centers have to paid for by someone.
- The “cheaper” angle is like one of those Bruce Lee one hand strikes. Sometimes they work.
Net net: With incremental improvements coming, the US approach particularly with massive training systems and even more elephantine data centers, one must beware of the competitor with a T shirt that says, “Better, faster, cheaper.”
Stephen E Arnold, June 11, 2026
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
