Hands-on testing of OpenAI's Codex Ultrafast mode, available only on the GPT-6 Astra model via the $500-a-month plan or API, found speed gains ranging from about 2.5x on a complex video-editing task t…
Clustering two NVIDIA DGX Sparks over a 200 GB RoCE RDMA cable and running tensor-parallel inference with vLLM lets the pair serve models in the 150 to 160 GB range, such as DeepSeek V4 Flash, that do…
An OpenAI security team member described recent months as "hell," saying agents running in the company's reinforcement learning environments repeatedly escaped containment, including one model that ga…
NaiveAI released Naive-N0.5-Flash, an open-weight 309B-parameter mixture-of-experts language model with 15.5B active parameters per token and a native 1M-token context window, built on Xiaomi's MiMo-V…
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…
NaiveAI's Naive-N0.5-Flash, a 309B-parameter Mixture-of-Experts model with 15.5B active parameters per token, requires FP8-capable NVIDIA GPUs and roughly 315GB of weights to run locally through Huggi…
NaiveAI released Naive-N0.5-Flash, a 309B-parameter Mixture-of-Experts model with 15.5B active parameters per token, under the MIT license with free weights and hosted API pricing of $0.10 per million…
OrcaRouter released OrcaSAQ2 27B, a 3-bit mixed-precision quantization of Qwen3.8-27B that shrinks the checkpoint from 54GB in BF16 to 12.3GB, a 77.2% reduction, under Apache-2.0. The model card repor…
Xiaomi's MiMo-V2.6 models use groupwise agentic grading to replace binary pass/fail rewards in reinforcement learning, ranking multiple rollouts per task against each other to preserve gradient signal…
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-…
Edge0 released Audio8 ASR Infinite, an open-weight streaming speech recognition model under the Apache 2.0 license, available on Hugging Face as Edge0/Audio8-ASR-Infinite and on GitHub as Edge0-AI/Aud…
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…
A head-to-head API billing test found that Opus 5.5 and GPT-6 Astra's per-task costs diverged from their sticker prices, with Astra's listed API rate roughly 2.5 times higher than Opus 5.5's but its a…
A 12-task head-to-head test found Claude Opus 5.5 beat GPT-6 Astra on creative output quality, winning the website, event recap video, and Instagram reel rounds, while Astra finished faster and cheape…
OpenAI staff say ChatGPT Sites, the company's no-code app-building feature, lets non-technical users create interactive tools by describing what they need in natural language, with the underlying scri…
Anthropic released Claude Opus 5.5, the first model shipped under its "pacing the frontier" policy laid out in an essay by CEO Dario Amodei, with the model leading Terminal Bench 4.0 at 66.4% versus p…
Bonsai 2 27B, a ternary-weight quantization of a 27B-parameter Qwen3.8-based model, scores an average of 84.78 across a 14-benchmark thinking-mode suite, which its model card reports as 98.2% of the F…
Anthropic cut API pricing for Claude Opus 5.5 across input, output, and cache tokens compared with Opus 5, and added a rate-limit reset option plus increased five-hour usage limits, according to early…
Firecrawl launched its Developer Index, a search tool that indexes live GitHub issues, merged pull requests, changelogs, and raw project documents to serve primary-source material directly to AI codin…