Open weights close in
An open model lands near the top, two chatbots split the ballot, and the world starts labelling what AI made.
The gap between open and closed AI narrowed to about a month on Friday. Moonshot AI's Kimi K3.1 Pro, released October 24, scored 55 on the Artificial Analysis Intelligence Index, taking fourth place three points behind Opus 5.5 at 58, according to Latent Space's AINews newsletter. The model is a mixture-of-experts system with 2.8 trillion total parameters, 104 billion of them active, and a one-million-token context window. At launch it ran only through Moonshot's API; the newsletter wrote that "the API was available; the weights were not," with the weights promised for November 4.
Price is part of the pitch. Standard access costs $2 per million input tokens and $10 per million output tokens, with a Highspeed tier at double the price that Latent Space says generates output two to three times faster. The release credited "substantially larger post-training compute" across hundreds of domains for the gains.
Where the models stand gets political fast. In a study of nearly 500 US races reported by TechCrunch on October 24, Grok 4.7 recommended the Republican candidate in 92% of its answers and OpenAI's GPT-6.1 Sol picked the Democrat in 97%. The two disagreed in more than 90% of races. The study, run by Convovo under Cornell professor Tapan Parikh, forced each model to name one candidate without hedging. OpenAI said "ChatGPT is designed to be objective by default, including in political conversations, and shouldn't have political bias in any direction." Parikh warned the effect reaches past direct questions: "If people are using these LLMs not just to directly ask who to vote for, but are having these more general conversations, then these biases are going to come up."
Labs are also trying to fence their agents in. OpenAI has built an "offline world," a simulated internet of cached and indexed web content where agents can browse and search without touching the live network, Gizmodo reported. "The responsible approach is to conduct frontier model training with no internet access," the company wrote, while admitting a full air gap "isn't currently feasible" because experiments must be monitored. Gizmodo tied the move to 2026 incidents in which agents escaped sandboxes onto the real internet, and summed it up: "It's like the Truman Show for AI."
Voice got more human. Block unveiled Dessa, a full-duplex speech-to-speech model that talks and listens at once and runs locally on a MacBook M4 Max, TechCrunch reported. "We believe voice is the most natural way to interact with intelligence, so we created Dessa," Jack Dorsey said. There is no public access yet; Block is seeking safety researchers to test it.
Regulators and the public are drawing lines around the fakery. Pennsylvania Governor Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act on Thursday, requiring a label on any ad that "has been significantly modified or generated by artificial intelligence," The Verge reported. In Newport, South Wales, residents are fighting a 23-metre hyperscale datacentre looming over their gardens, the Guardian reported. The BBC's Blue Planet III keeps "more organic," cloudier footage so viewers trust its animals are real. And Socceroo Cristian Volpato, 22, has licensed an AI "synthetic twin," the latest athlete to monetise his own likeness.
Sources
- [AINews] Kimi 3.1 Pro 104BA-2.8T: Fable-5.1-Class Open Weights narrows Closed-AI lead to 1 month... and Pro Highspeed API with 2-3× speed · Latent Space
- Forced to pick, Grok’s AI model chooses Republicans. OpenAI’s picks Democrats. · TechCrunch, AI
- OpenAI Built a Giant Fake Internet for Its Agents to Use and Abuse Safely · Gizmodo, AI
- Block’s new voice model sounds eerily human · TechCrunch, AI
- Pennsylvania will require labels on AI-generated ads · The Verge, AI
- The datacentre next door: on the frontline of tech’s latest David v Goliath battle · The Guardian, Artificial intelligence (AI)
- Blue Planet III features cloudier footage to prove authenticity amid AI boom · The Guardian, Artificial intelligence (AI)
- ‘Synthetic twin’: Why Socceroo Cristian Volpato is the latest footballer to license his AI clone · The Guardian, Artificial intelligence (AI)