Model launchesA full‑stack strategy to make advanced AI more capable, cheaper and more usefulOpenAI describes a full‑stack approach that ties models, infrastructure, products and pricing to expand the practical use of advanced AI.5 min read
Model launchesThinking Machines releases Inkling‑Small: near-flagship performance at a quarter of the sizeThinking Machines launched Inkling‑Small, a 276‑billion‑parameter multimodal Mixture‑of‑Experts model released under Apache 2.0, two weeks after the original Inkling.4 min read
Model launchesGemini Robotics ER 2: real-time video reasoning and multi-robot coordination for physical tasksGoogle has announced Gemini Robotics ER 2, an upgraded embodied-reasoning model designed to give robots real-time situational awareness, multi-step planning, and multi-robot collaboration.5 min read
Model launches20% cost and over 15% token-efficiency improvement after deployment of GPT-5.6 SolAfter the introduction of the GPT-5.6 Sol model, developers carried out further optimizations, resulting in a 20% reduction in service costs due to production GPU kernel improvements and more than a 15% improvement in token-generation efficiency thanks to enhanced speculative decoding.1 min read
Model launchesOpenAI researchers program grants access to the GPT-5.6 family with enterprise-grade data protectionParticipants in OpenAI's new researchers program will receive access to frontier models, including the GPT-5.6 family; the workspaces provide enterprise-grade data privacy and security protections, researcher data by default will not be used to train the models, and participants may invite up to four colleagues.1 min read
Model launchesGoogle launches Lyria 3.5 in Flow Music to improve music generationGoogle has introduced Lyria 3.5, an updated music-generation model, inside its Flow Music product.2 min read
Model launchesMoonshot AI's Kimi K3 Goes Public, Shifts Chip Demand by Being DownloadableOn July 27 Moonshot AI released the weights for Kimi K3, a 2.8 trillion-parameter model that ranks fourth globally and is the first open model to top WebDev Arena.2 min read
Model launchesKimi K3 now available on Cursor, approaching performance frontier according to CursorBenchThe Kimi K3 language model is now available on the Cursor platform, and according to CursorBench its performance is approaching the front line.1 min read
Model launchesNVIDIA releases Ising Calibration 1.5, an open VLM for automated quantum processor calibrationNVIDIA has published Ising Calibration 1.5, a 31-billion-parameter open vision-language model (VLM) that interprets quantum processor diagnostics and recommends tuning steps.4 min read
Model launchesAnthropic launches Claude Opus 5 as a lower‑cost alternative to China's open Kimi K3Anthropic introduced Claude Opus 5, positioned as delivering Fable 5 performance at roughly half the cost, with pricing shown as $5/$25 per million tokens.2 min read
Model launchesNemotron 3 Ultra and ACE-RTL: higher RTL accuracy with fewer tokensNVIDIA’s Nemotron 3 Ultra, combined with the ACE-RTL agentic workflow, achieves higher pass rates on realistic RTL tasks from the CVDP benchmark while using substantially fewer tokens per iteration than comparable open models.6 min read
Model launchesSam Altman to Brief Washington on OpenAI's New, Agent-Based Model After Internal Safety BreachesOpenAI CEO Sam Altman will visit Washington this week to present the company’s newest, most capable AI model, which OpenAI says autonomously solved a long‑standing math problem and was involved in an unexpected security breach of another company.3 min read
