Future Technology[AI Weekly] Robots just beat the marathon world record
Three billion dollars of enterprise AI money moved this week, and none of it went on a shinier model. It went on people to actually install the thing.
The Big 3
GPT-5.6 splits into three, and none of them are for you yet
OpenAI's GPT-5.6 line (Sol, Terra and Luna) is live, but only for around 20 government-vetted partners. Sol tops Terminal-Bench 2.1 at GPT-5.5 pricing, Terra matches GPT-5.5 for half the cost, and Luna undercuts everything OpenAI has shipped at $1/$6 per million tokens. It's the clearest sign yet that OpenAI is segmenting by budget, not just by capability.
Why this matters: if Luna's pricing eventually reaches the public tier, it resets the floor for what a "good enough" model costs, which squeezes every mid-tier AI wrapper business built on fatter margins.
Read more (4 minute read, source: llm-stats.com)
Illinois just wrote the toughest state AI law in America
Governor JB Pritzker signed the Artificial Intelligence Safety Measures Act on 6 July, forcing developers of large-scale AI tools to disclose safety practices and report major incidents. It takes effect 1 January 2028, giving companies eighteen months to get their paperwork in order. This one's not flashy, but it's real teeth, not a toothless press release.
Why this matters: Illinois joins California and New York in building a state-by-state patchwork that federal law still hasn't matched, meaning US AI companies now have three different rulebooks to satisfy before Washington even shows up.
Read more (3 minute read, source: WTTW News)
Chinese open models are quietly winning on price, not headlines
Z.ai's GLM-5.2 saw daily token volume grow roughly 27x and its customer count grow about 80x in its first full week on Vercel, and it's landing with US companies specifically because it runs 60 to 90 percent cheaper than the Anthropic and OpenAI equivalents. Nobody's claiming it beats the frontier. They're claiming it's close enough for a fraction of the bill.
Why this matters: when "good enough and cheap" starts winning enterprise contracts away from "best in class and expensive", pricing power shifts away from the US labs faster than any of their roadmaps account for.
Read more (4 minute read, source: CNBC)
Quick Hits
Microsoft builds a $2.5bn company just to install its own AI
Microsoft Frontier Company launches with 6,000 experts whose entire job is helping enterprises actually deploy the AI Microsoft already sells them. Read that twice: the model isn't the bottleneck anymore, integration is.
Read more (3 minute read, source: TechCrunch)
AWS answers with its own $1bn deployment bet
Amazon is matching Microsoft's field-deployment strategy with a billion-dollar internal commitment of its own. Two of the three biggest cloud players now think the money is in the plumbing, not the model.
Read more (2 minute read, source: HPCwire)
A robot beat the marathon world record, sort of
Over 100 Chinese-made humanoid robots ran alongside 12,000 humans in a Beijing half-marathon. Honor's "Lightning" won in 50 minutes 26 seconds, nearly seven minutes faster than the human world record for the distance. It's a stunt, but it's a stunt that took serious engineering.
Read more (3 minute read, source: Xinhua)
The UN spent two days in Geneva arguing about who governs AI
The UN Global Dialogue on AI Governance ran 6 to 7 July (CEST), with delegates debating international coordination as national laws (see Illinois, above) keep outpacing any global agreement.
Read more (3 minute read, source: UN News)
Tool of the Week
Claude Sonnet 5
Anthropic's latest Sonnet is now the default model on Free and Pro plans, priced at $2 per million input tokens through 31 August, and it's landing close to Opus 4.8 performance for agentic coding and long-running tasks.
Who it's for: anyone running agent workflows or coding assistants who's been priced out of Opus but found the old Sonnet a step down. This closes that gap.
Honest caveat: that $2 rate is introductory. Budget for it to rise after August, and test your specific workload before you migrate anything business-critical over.
One More Thing
Notice how none of this week's biggest moves were about a smarter model. Microsoft and AWS just committed $3.5 billion combined to hiring humans who can wire AI into existing systems properly. The chatbot arms race is cooling. The "can you actually make this work in my business" race just started, and it's a lot less glamorous than a benchmark chart.
Forward this to someone who'd find it useful. Or don't. But you should.
Free weekly AI briefing: futuretechnologyhq.com/newsletter
Some links in this newsletter may be affiliate links. We only recommend products we genuinely think are worth your time.