Live signal feed

Laguna M.1/XS.2 Technical Report

New MoE models Laguna M.1 (225B params) and XS.2 (33B) for coding tasks. Could be relevant if they release APIs.

ArXiv cs.AI

A Simple State Space Model Excels at Multivariate Time Series Classification

State space models beat Mamba for time series classification. Could be useful if you're building forecasting tools, but limited scope.

ArXiv cs.LG

ElevenLabs’s new music generation model can switch genres mid-track

ElevenLabs launches music generation with mid-track genre switching. Could be useful for solo devs building content creation tools.

TechCrunch AI

The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence

MiniMax-M2: 230B parameter MoE model, only 9.8B active per token. Built for agentic deployment.

ArXiv cs.AI

AirCast-SR: A Foundation Model for Kilometer-Scale Atmospheric Super-Resolution via Latent Consistency Diffusion

Weather super-resolution foundation model scales 28km forecasts to 1km detail. Useful for location-based apps needing weather data.

ArXiv cs.LG

InfoQuant: Shaping Activation Distributions for Low-Bit LLM Quantization

New quantization method shapes activations for better low-bit LLM performance. Could reduce inference costs if implemented in your model hosting.

ArXiv cs.LG

Max-Window Scale Estimation for Near-Lossless HiF8 W8A8 Quantization-Aware Training

HiF8 quantization method for efficient LLM deployment with minimal quality loss. Could significantly cut your inference costs if available.

ArXiv cs.LG

GAC: Noise-Aware Adaptive Mixing for Hybrid SFT-RL Post-Training

Adaptive mixing for hybrid fine-tuning that adjusts to noise levels. Useful if you're training custom models with mixed supervision signals.

ArXiv cs.LG

modelscope/FunASR

FunASR: Industrial speech recognition 170x faster than Whisper, 50+ languages, streaming API. Serious Whisper alternative for voice apps.

GitHub Trending Python

Warp’s big bet on building open source with GPT-5.5

Warp using GPT-5.5 for coding agents across dev workflows. First mention of GPT-5.5 - suggests new model capabilities for code coordination.

OpenAI Blog