Cantina Security, with Yeta Labs, released apex-flash-1, an open-weights vulnerability-research model built as a reinforcement learning fine-tune of Z.ai's GLM-5.3-Flash, solving 40 of 60 held-out bug…
IQuest released IQuest-Q1, an open-weight 320B-parameter Mixture-of-Experts coding model that activates only 15B parameters per token and supports a 524,288-token context window, with official deploym…
Janus, an MIT-licensed single Go binary from the Vibra-Ingenn project, launched as a Show HN on October 1 with 42+ points, wrapping llama.cpp's Vulkan backend to run GGUF models on AMD, Intel, or NVID…
A paper submitted to arXiv on 29 Sep 2026 introduces Context Language Models (CLMs), language models that natively manage their own context by treating the context as a file the model can update witho…
Researchers submitted Purlin, a scale-up GPU collective communication framework that separates orchestration from the datapath, to arXiv on 29 Sep 2026. Purlin specifies collectives as input/output la…
IQuest released IQuest-Q1, an open-weight 320B-parameter Mixture-of-Experts language model with 15B active parameters per token and a 512K-token context window, built for agentic coding and multi-step…
IQuest released IQuest-Q1, a 320-billion-parameter Mixture-of-Experts model with roughly 15 billion active parameters per token, an 88-layer transformer, 256 experts (8 active), and a 524,288-token co…
UniServe, a serving engine for FastH3 8-Step text-to-video-with-audio generation from the hao-ai-lab, delivers lower median end-to-end latency and 20–46% higher throughput than FastVideo, vLLM-Omni an…
INT21 generated 20 inference engines across seven model categories in two weeks using Rust with C++ and CUDA components, and its MiMo engine reached 1,308 tokens/s versus 540 for tuned SGLang and 1,01…
Enterprise data breaches involving unsanctioned Shadow AI cost organizations an average of $650,000 more than conventional security incidents, and 20.0% of global enterprises have already suffered a p…
Sference implemented prefill concurrency in SGLang to improve time to first token (TTFT) in multi-tenant deployments, where stock SGLang allows only one request to be mid-chunk at a time and that requ…
Xiaomi released MiMo-V2.6-Flash-RL, a sparse Mixture-of-Experts omnimodal model with 309 billion total parameters and 15 billion active per token, positioned as the efficiency-focused sibling to MiMo-…
Liquid AI released LFM2.5-VL-3B-DSpark, an experimental speculative-decoding draft model for its LFM2.5-VL-3B vision-language model that adds about 279.5M parameters and delivers up to 3.13x faster de…
Researchers introduced ServingStudio, an integrated workbench that combines a Rust-based LLM serving simulator with an agent to automate performance profiling and optimization. The ServingStudio Simul…
Xiaomi's MiMo team released MiMo-V2.6-Pro-RL, an open-weight trillion-parameter mixture-of-experts agentic model with 1.02 trillion total parameters, 42 billion active per token, and a 1 million token…
BottleCap AI released ThinkingCap-Qwen3.8-27B, a fine-tune of Qwen3.8-27B that cuts thinking tokens by an average of 37.2% across 12 benchmarks while macro-average accuracy falls 0.86 percentage point…
LiquidAI released LFM2.5-VL-DSpark, a 279.5M-parameter speculative decoding drafter for its LFM2.5-VL-3B vision-language model that adds 8.9% to the target model's parameter count while delivering dec…
Nunchux generated a 5-second MiniMax-H3 video in 1.33 seconds on a single server with eight AMD MI355X GPUs, a 21.8× speedup over SGLang on the same hardware, according to the company's benchmark post…
Researchers introduced SWE-Serve, a benchmark of 53 repository-grounded tasks derived from recent production changes to SGLang, spanning six inference engineering families, to evaluate agents on produ…
Oh My Pi (omp) 18.2.7 changed how custom local models are configured, requiring users to rename the provider from `local` to an unused name and add `qwenTemplateReasoningEffort: true` to a model's `co…