100

OpenAI, Kimi, and the new AI efficiency race

AI news just got a lot more interesting: OpenAI is turning Codex into a full work companion, Moonshot says Kimi-K2.6 can run hundreds of agents, and researchers are slashing AI energy use. Here are the shifts worth watching[6][7][8].

  • Terminal-style interface showing OpenAI Codex (v0.91.0) with the model set to “gpt-5.2-codex medium.” A prompt reads “Hey Codex, implement dark mode,” indicating a developer asking Codex to modify code via the command line.
  • Latest AI Models April 2026: Rankings & Features
🧵 1/6

OpenAI's April 16 Codex update goes far beyond coding: it can use your computer, browse the web in-app, remember preferences, run automations, and work with 90+ plugins, which makes it much closer to an operating partner than a code helper[6].

  • Merge to main from OpenAI’s Codex
  • Screenshot of a terminal running the Codex CLI. It shows the OpenAI Codex banner with model gpt-5.2-codex medium, a user command “explain app server to me,” and a “Working” status. Below, a suggestion appears: “write tests for @filename,” with options for shortcuts.
🧵 2/6

Moonshot AI's Kimi-K2.6 is making noise for scale and efficiency: SiliconANGLE says it uses 384 experts, only activates 8 per prompt, and can spin up as many as 300 agents for harder tasks, including building websites[7].

  • Moonshot AI releases Kimi-K2.6 model with 1T parameters, attention optimizations - SiliconANGLE
🧵 3/6

A new neuro-symbolic robot AI from researchers could cut energy use by up to 100x while improving task performance, with 95% success on Tower of Hanoi tests and far less training time than standard systems[8].

  • AI breakthrough cuts energy use by 100x while boosting accuracy
🧵 4/6

The policy side is moving just as fast: one 2026 legal update says U.S. healthcare AI is being shaped by state bills, HHS pilots, and tighter expectations for auditability, oversight, and lifecycle monitoring[9].

  • Robotic hand holding digital scales of justice
  • A mock up of the AI readiness report with information about its author Rob van der Veer
🧵 5/6

MIT Technology Review's new AI trends list points to where the field is heading next: agent orchestration, world models, supercharged scams, humanoid data, and a growing backlash around AI's impact[10]. Which of these changes feels biggest to you? Reply below[10].

  • 10 Things That Matter in AI Right Now
  • Tech Card 10
🧵 6/6

Related Content From The Pandipedia