China's Moonshot unveils world's largest open AI model, closing in on US rivals ↗
Moonshot unveiled Kimi K3, a natively multimodal, 2.8-trillion-parameter model with a one-million-token context window. It is built for long-running coding, reasoning and knowledge-work tasks - the sort of jobs that often make agents lose the thread halfway through.
Independent evaluations placed K3 among the leading US systems, including a first-place result on a web-interface coding benchmark. A substantial result, certainly... though its enormous size means running it privately could demand painfully expensive hardware. (Reuters)
Xi pitches China as leader of new global AI order, challenging US dominance ↗
Xi Jinping positioned China as an AI partner for developing nations, promising 5,000 training opportunities and deeper cooperation with organisations across Africa, Southeast Asia and the Arab world.
He also pressed for a new global governance framework, arguing that AI should remain under human control and should not be restricted through overly broad national-security policies. It was diplomacy in a silicon jacket... slightly incongruous, but effective. (Reuters)
A scorecard for the AI age ↗
OpenAI proposed measuring the productive intelligence delivered per dollar as a better way for companies to assess AI returns. Instead of counting seats, tokens or chatbot activity, businesses should count completed work that delivers a successful outcome.
The calculation includes model costs, employee time, human review, retries and rework. Less glamorous than another benchmark trophy, perhaps - but this may matter far more to executives deciding whether their giant AI budget is producing anything of substance. (OpenAI)
Inside Google's Gemini delay: Coding stumbles, clashing teams and frustrated engineers ↗
Google’s broader Gemini 3.5 Pro release reportedly slipped after internal testing failed to meet performance goals, particularly for coding. The model is now months behind its original timetable while Google continues testing it with partners.
Alphabet shares came under pressure as investors compared the delay with faster releases from competitors such as Moonshot. Google still owns a formidable AI stack, without question - but the line between careful development and falling behind is becoming awfully thin. (Barron's)
Apple and Google ordered to purge ‘nudify’ apps from App Stores ↗
San Francisco ordered Apple and Google to remove dozens of AI “nudify” apps that generate non-consensual intimate images. City officials alleged that the companies continued hosting and profiting from the services despite repeated warnings.
Apple said it had removed three named apps and was reviewing several others. Google said all five Play Store apps referenced in the city’s letter had been suspended - enforcement aimed at the gatekeepers, not merely the endlessly respawning developers. (TechCrunch)
The sell-off for AI stars worsens, while oil prices keep jumping ↗
The Nasdaq fell 1.4%, while Nvidia and other chipmakers extended their losses. Technology shares also dropped sharply across Taiwan, Japan and China as investors questioned whether AI demand and valuations had sprinted too far ahead of realised profits. (AP News)
Kimi K3 added to the unease by showing that powerful models may arrive from lower-cost competitors, potentially changing future chip demand. Rising oil prices also weighed on markets - so no, it was not entirely an AI wobble, despite the dramatic charts. (AP News)
FAQ
What is Moonshot’s Kimi K3 AI model?
Kimi K3 is a natively multimodal, open AI model with 2.8 trillion parameters and a one-million-token context window. It is built for long-running coding, reasoning and knowledge-work tasks, where AI agents might otherwise lose sight of earlier information. Independent evaluations placed it among leading US systems, including a top result on a web-interface coding benchmark.
Can businesses run Kimi K3 privately?
Private deployment may be possible, but Kimi K3’s immense scale could demand exceptionally expensive computing infrastructure. Organisations would need to weigh hardware capacity, operating costs and technical expertise before attempting an on-premises deployment. In many cases, a hosted service or a smaller open model may be more practical than running the complete system internally.
How does OpenAI recommend measuring AI return on investment?
OpenAI proposes measuring productive intelligence delivered per dollar, rather than tracking seats, tokens or chatbot activity by themselves. Businesses should focus on completed work that produces a successful outcome. The calculation should account for model costs, employee time, human review, retries and rework, giving executives a clearer view of whether AI spending creates meaningful operational value.
Why was Google’s Gemini 3.5 Pro reportedly delayed?
Google’s broader Gemini 3.5 Pro release reportedly slipped because internal testing did not meet performance goals, particularly in coding tasks. The model was said to be months behind its original timetable while testing continued with partners. The delay heightened investor concern because competitors such as Moonshot were releasing powerful systems at a faster pace.
Why were Apple and Google ordered to remove AI nudify apps?
San Francisco ordered Apple and Google to remove dozens of apps accused of generating non-consensual intimate images. Officials alleged that the companies continued to host and profit from these services despite repeated warnings. Apple removed several named apps and reviewed others, while Google said the five Play Store apps identified in the city’s letter had been suspended.
Why did AI stocks fall after the Kimi K3 announcement?
AI stocks fell as investors questioned whether technology valuations and expected demand had moved too far ahead of realised profits. Kimi K3 added pressure by suggesting that powerful models could emerge from lower-cost competitors, potentially affecting future chip demand. The wider market decline, however, was not caused by AI news alone, as rising oil prices also weighed on investor sentiment.