falproduct update
fal introduces interactive Worlds platform
Introduced fal Worlds on October 15 for creating and interacting with generated scenes using available world models.
What changed in AI.
Market developments ordered by event date, with direct evidence. Site maintenance has its own publishing log.
Market developments · 181
Krea AImodel release
Krea releases Krea 3 image editing model
Released Krea 3 on October 15 for native-2K image editing through Krea and partners; ComfyUI provides distribution, not model ownership.
Microsoftresearch
Microsoft reports MAI-Base-2 pre-training complete
Reported completed pre-training of MAI-Base-2 on October 15; the model was still in post-training, with no public weights or general availability announced.
Nvidiamodel release
NVIDIA releases Nemotron 3.5 Super VL and Bolt models
Released Nemotron 3.5 Super VL and Bolt 74B-A9B on October 15; the E4B 4GB variant was announced for October 23, not yet available.
OpenAImodel release
OpenAI begins GPT-Live-1.1 voice rollout
Began rolling out GPT-Live-1.1 and Quip voice models October 15 in ChatGPT, Codex and Dots; API access was still planned.
Origincompany
Origin emerges with OG-1 and $20M total funding
Origin disclosed $20M cumulative funding on October 15 and described deployed OG-1 robots for wall finishing; the amount is not mapped as a new standalone financing round.
Perplexityproduct update
Perplexity open-sources SPACE sandbox
Open-sourced SPACE, an agent-task sandbox used by Perplexity Computer, under Apache 2.0 on October 15.
Sarvam AIproduct update
Sarvam introduces Voice Agents MCP
Introduced an MCP integration for building Sarvam voice agents and working with campaigns and calls from coding tools on October 15.
Vercelproduct update
Vercel Sandbox Drives reaches general availability
Made Sandbox Drives generally available on October 15, advancing the same persistent-storage product from its September public beta.
Appleresearch
Apple details distributed MLX inference
Published MLX research on October 15, 2026 describing updated open-source support for local and distributed inference across Mac Studios using RDMA over Thunderbolt 5. Demonstrated third-party models are not Apple Foundation Models.
Meta AImodel release
Meta releases Muse Spark Open
Released Muse Spark Open on October 15, 2026, an open-weight model under the Muse Model License, with API and Muse Code access. Open weights do not imply an open-source license.
Mistral AImodel release
Mistral releases Timestral forecasting models
Released Timestral on October 15, 2026, a time-series foundation-model family in four sizes for probabilistic multivariate forecasting, with a public repository and Mistral Forge adaptation options.
OpenAIcompany
Arkansas attorney general files suit against OpenAI
The Arkansas attorney general filed a lawsuit on October 14 alleging deceptive ChatGPT safety representations and child privacy violations; the allegations remain unadjudicated.
Nvidiaproduct update
NVIDIA introduces cuPotential early access
Introduced cuPotential, a CUDA-X library for atomic and molecular simulation agents, in early access on October 14; reported speedups are from NVIDIA early-access testing.
HeyGenresearch
HeyGen introduces scene-aware voice model
Introduced HeyGen Voice on October 14, 2026, a scene-aware voice-cloning speech model for talking videos, with a product trial linked from its research page. A separate avatar model renders the video.