Search “Speech”

Google DeepMind Releases Genie 4 Real-Time World Model with Native Audio Generation

Google DeepMind has released the Genie 4 world model as a research preview to trusted testers; it can generate explorable worlds in real time from text and image prompts, and natively generate ambient sound, music, and character dialogue. The model introduces Gemini 3.8 Flash as a planner within the simulation loop, responsible for breaking down goals, inspecting frames, and replanning, which DeepMind calls an early step toward hybrid world models. Genie 4 is hosted only on Google's platform and will later be opened to AI Ultra users and Gemini API partners; all video and audio carry SynthID watermarks.

Liquid AI Releases Six 2B-Class Open-Weight Models in the LFM3 Series

On October 27, Liquid AI released the LFM3 series. All six 2B-parameter-class open-weight models branch from the same LFM3-1.7B-Base, covering text, vision, audio, omni-modal, and decision models. LFM3-1.7B has 1.73B parameters, a 256K token context, supports 16 languages, and adopts a Tri-Hybrid architecture with 13 layers of gated short convolutions, 5 layers of sliding-window attention, and 6 layers of global attention. It is available under the LFM Open License v1.0 on Hugging Face, the Liquid AI Playground, and via API. The company claims that LFM3-1.7B surpasses LFM2.5-2.6B with 35% fewer parameters and occupies about 1 GB of memory at 4-bit precision, but the model card states that it is not suitable for agentic coding or complex multi-step reasoning.

Liquid AI Releases the LFM3 Series of Models, Including 1.7B Text, Vision, Audio, and Omni Versions

Liquid AI has released the LFM3 series of open-weight models, all branched from LFM3-1.7B-Base, covering text, vision, audio, Omni, and the d1 decision model. LFM3-1.7B-Base adopts a three-hybrid-layer architecture, is pretrained on 22T tokens, and in the mid-training stage over the final 3.5T tokens incorporates more agent trajectories and tool-calling data, with the context window extended to 256K tokens. The company says LFM3-1.7B is about 25% faster than LFM2-1.2B at an 8K context, and scores 38.42 versus 3.02 on WebArena-Verified; LFM3-VL-2B's average score across 62 vision benchmarks rises from 57.6 to 71.4. All models are available for download on Hugging Face, along with a single-GPU fine-tuning notebook.

Google DeepMind releases world model Genie 4, supporting native audio and real-time interaction

Google DeepMind has released the world model Genie 4, which can generate interactive worlds in real time and, for the first time, natively generate ambient sound and character dialogue. The model builds on capabilities from the Gemini series; Gemini 3.8 Flash breaks down goals in a loop and checks frames to maintain causal coherence across multi-step events, and Gemini 4 Argon has also been used for joint simulation tasks with Gemini Robotics. The company says it remains an early experimental model: over long runs, object permanence and complex physical interactions still drift, all generated content carries a SynthID watermark, and it is currently available in preview to trusted testers first.

Moonshot AI releases Kimi K3.1 Pro: 2.8T total parameters, 104B activated parameters, weights scheduled to be opened on November 4

Moonshot AI releases Kimi K3.1 Pro, a multimodal MoE model with 2.8T total parameters, 104B activated parameters, and a 1 million token context window. The Standard and Highspeed versions are now available on the Kimi website, App, Kimi Work, Kimi Code and API. The company says Highspeed offers 2–3x faster output at twice the price of the Standard version. Artificial Analysis ranks it fourth on the Intelligence Index with a score of 55, behind Opus 5.5's 58; on Arena's Code Arena WebDev it ranks 8th with 1692 points, up nine places from K3. Standard pricing is $2 per million tokens for input, $10 for output, and $0.2 for cached input, one-third lower than K3; the weights have not yet been released, and the company promises to open them on November 4.

NVIDIA Expands Nemotron Model Family with New Super VL and Two Bolt Models

NVIDIA has added three models to the Nemotron open-source model family, covering enterprise workflows, local AI agents, and conversational applications across data centers, desktops, and edge devices. These include the 120-billion-parameter Nemotron 3.5 Super VL, which NVIDIA claims delivers up to 5x the throughput of the next-best comparable model, with accuracy approaching that of its largest reasoning model at roughly one-fifth the parameter count; there are also two small models in the Nemotron 3.5 Bolt series: 74B-A9B 24GB for everyday agent tasks on RTX and DGX Spark, and E4B 4GB for conversational and physical AI agents. Super VL and Bolt 74B-A9B are already available, and Bolt E4B is scheduled for release on October 23.

Hugging Face Releases Beam: A 501B Open Source Model Supporting 1M Context and Self-Hosting

Hugging Face has released Beam, a 501B-parameter sparse MoE open source model aimed at long-text reasoning and agent tasks with 1M context, activating only about 23B parameters per token. The model uses 52 layers of hybrid attention, with 43 layers using a 512-token sliding window and 9 layers using global attention; the KV cache grows by 36 KiB per token, about one-sixth that of a full-attention approach. The official release provides three checkpoints—BF16, FP8, and NVFP4; NVFP4 can compress the weights from about 1 TB to about 280 GB, and day-0 support is already available in transformers, vLLM, and SGLang.

English translation is being prepared.

Swiss AI releases the Apertus 1.5 multimodal model, expanding context to 262k tokens

Swiss AI has released Apertus 1.5, which builds on 1.0 by adding native image and audio understanding, a thinking mode, and a 262,144-token context window—4 times larger than 1.0. The model uses early fusion to convert images and audio into discrete tokens and feed them into the same sequence as text; the architecture is unchanged, and it only understands multimodal input while outputting text only. The team says the 8B version outperforms open-source models of the same scale on instruction following and multilingual tasks, its multimodal performance is close to Gemma 4 12B, and it has also released the training data mixture, preference data, and intermediate checkpoints from each stage.

GPT-6 Family's First Model Astra Released, Author's Hands-On Test Consumed 4 Billion Tokens

OpenAI released Astra, which the author calls the beginning of the GPT-6 family; it performs outstandingly on ARC-AGI-3 and scored highest on Zapier's AutomationBench, with pricing the same as Fable 5.1. During weekend trials, the author consumed 4 billion tokens and found its capability fluctuates noticeably—it is a showcase model rather than one that reliably does work. Astra in Codex can now continue advancing tasks without waiting for an answer, and OpenAI also claims it has achieved its automated research intern goal, with the next goal being an automated AI researcher by March 2028.