ByteDance Seed and Tsinghua AIR released CUDA Agent, an agentic reinforcement learning system that trains a large language model to generate CUDA kernels outperforming a compiler. On KernelBench, the …
MiniMax released MiniMax-Music3, an open-weights text-to-music model that generates complete songs of up to five minutes from lyrics and a structured caption, outputting 32 kHz, 16-bit stereo WAV. The…
DeepSeek released DeepSeek Harness v0.1 in developer preview, publishing the full source code under the MIT license at deepseek-ai/deepseek-harness. The harness, built on the Cordis plugin framework, …
A technical tutorial demonstrates an end-to-end supervised fine-tuning pipeline for the XYZ-Aquila-SFT dataset using Hugging Face Transformers, PyTorch, and PEFT, culminating in LoRA fine-tuning of Qw…
Z.ai released GLM-5.3 on August 14, 2026, reusing the 743B GLM-5.2 base model unchanged and achieving gains solely through scaled post-training. Terminal-Bench 3.0 scores rose from 4.6 to 28.3, DeepSW…
Cactus Compute released Needle 2, an open 45M-parameter tool-calling model that ships as a 14MB binary and runs a full session in about 28MB of RAM, targeting devices with no GPU or NPU. The model ach…
MarkTechPost published a tutorial on building a compact reasoning-focused language model by streaming the SupraLabs reasoning corpus from Hugging Face, applying quality filters, and fine-tuning SmolLM…
Google released Gemini 3.7 Flash, an AI model with improved reasoning, achieving 43.6% on FrontierCode 1.1 Main (up from 34.4%) and 65.3% on DeepSWE v1.1, available via API and enterprise at $0.75 per…
Liquid AI released LFM2.5-VL-3B, a 3.1B-parameter vision-language model for on-device deployment, averaging 69.4 across 28 vision benchmarks, matching InternVL-3.5-4B and 0.7 points behind Qwen3.5-4B.…
SpaceXAI released Grok 4.6, a post-training upgrade over Grok 4.5 that scores 61 on the Artificial Analysis Intelligence Index, up five points and tied with GPT-5.6 Sol Max, with a 500,000-token conte…
AllenAI's Open Instruct framework was used to build an end-to-end post-training pipeline for a compact instruction-tuned language model, integrating Supervised Fine-Tuning, Direct Preference Optimizat…
NVIDIA released Nemotron 3.5 Lightning, a 30B mixture-of-experts open model with 3B active parameters, and NeMo Switchyard, an open source routing library, to speed up high-volume agentic tasks. The m…
Xiaomi Inc.'s MiLM Plus team released PROVE, a perception-aligned evaluation framework for object removal, at ACM MM 2026, featuring two metrics—RC-S for spatial coherence and RC-T for temporal consis…
LTX released LTX-2.5, an open weights world model for video generation optimized for local inference on NVIDIA RTX GPUs and NVIDIA DGX Spark, cutting VRAM requirements so creators can run a frontier m…
WebAI released TwIL-LM, a two-model family of formal-logic reasoners at 1.7B and 3B parameters, with the 3B model TwIL-LM3 outperforming gpt-oss-120b on four of five formal-reasoning lanes while runni…
Comfy-Org released MiniMax-H3, a multimodal video and audio generation model, and a tutorial demonstrates implementing an end-to-end pipeline using ComfyUI as a headless inference backend. The pipelin…
Meta released Muse Glimmer, a 30-billion-parameter multimodal agentic model distilled from Muse Spark, under the Apache 2.0 license, designed to run on a single consumer GPU or Mac with no network cal…
ByteDance's Seed team has introduced SeedRealtime, a native audio-visual full-duplex LLM that fuses audio, video, and text in a single end-to-end architecture, enabling real-time multimodal interactio…
NVIDIA released NemotronLabs VoiceChat 11B, an open 11B end-to-end speech-to-speech model for real-time full-duplex conversation, achieving 448 ms smooth turn-taking latency on Full-Duplex-Bench 1.0 a…
The LLM observability platform market is projected to grow from $1.97 billion in 2025 to $2.69 billion in 2026 and reach $9.26 billion by 2030 at a 36.2% CAGR, according to The Business Research Compa…