Lead story
Why AI keeps picking 17: We tested randomness across 12 models
YFarmX tested 12 language models across number ranges, five languages and coin flips. Their answers reveal strong favourites and an unusually tidy idea of randomness.
AI Hub
Explore the future
of artificial intelligence
Models, agents, infrastructure and policy: what actually changed, and what it means.
- 64
- AI Index records
- 135
- AI explainers
- 91
- AI stories published
- 33
- verified events ahead
NVDANVIDIA$230.36up0.84%
MSFTMicrosoft$499.7down2.04%
GOOGLAlphabet$338.46down1.17%
AMDAMD$477.57up4.69%
Nasdaq · 05 Sept 2026, 02:07 UTC · refresh delayed
Deal watchNvidia signed the definitive agreement to buy Hugging Face at $12.93bn all told
State of play
TIME-SENSITIVEverified 10 September 2026
LLMsOpenBMB’s MiniCPM5-2B brings a 2.52-billion-parameter, Apache-2.0 model to local assistants and tool use. Its model card specifies a 131,072-token context; the early-September frontier releases remain in the dated log below.Image GenerationChatGPT Images 2.5 adds sketch-guided creation, image comments and more consistent editing. OpenAI reports up to 50% lower generation latency than Images 2.0; Flare and Sunburst provide two API options. Older arena rankings below remain dated snapshots.Video GenerationNVIDIA’s Sol-H3 research reports a 1.653-second warm run for about five seconds of video with stereo audio on eight B300 GPUs. The four-step result is a hardware-specific inference benchmark, with model loading and final MP4 encoding outside the timing.
Open AI State of Play: all 7 areas Explore the AI desk
Learn the ecosystem
AI explainers
Open a field to explore the models, assets and machines behind the news.
- 58 guidesLarge Language ModelsThe models that read, reason and write, from the frontier labs to the open-weight field.
- 12 guidesRoboticsMachines that act in the world: the models that drive them, the bodies they drive, and the rules around both.
- 15 guidesAgentic AIThe autonomous agents that run tools, write code and act on your behalf, not just answer.
- 8 guidesWorld ModelsSystems that simulate interactive, explorable environments a frame at a time.
- 21 guidesVideo GenerationText and image to video models, from the safe Western defaults to the leaderboard leaders.
- 14 guidesImage GenerationText to image and image editing models, from frontier quality to the open-weight stack.
- 7 guidesAudio and VoiceMusic, speech and real-time voice models, and the licensing questions they raise.
Coming up
See the AI calendar →15 Sept to 17 SeptDreamforce 2026Salesforce's annual conference takes over San Francisco for three days, with recent editions centred on AI and the company's Agentforce platform.San Francisco, USA23 Sept to 24 SeptMeta Connect 2026Meta's annual developer conference on AI and mixed reality, where the company shows its latest models, glasses and headsets. Livestream registration is open.Online / Menlo Park, CA
Listen: AI briefings
All briefings →08 Sept 2026 · 00:01:38MiMo takes the desktopMiMo Desktop's invite beta brings document creation, live previews and browser control. Xiaomi's access page currently excludes the UK and EU.08 Sept 2026 · 00:01:59A fluid proof under reviewOpenAI has released a forced Navier-Stokes blowup construction and Lean formalisation, alongside a dispute over research priority and private data.