Artificial intelligence, professionally covered
Every consequential model release in one place: frontier launches, open-weight drops, capability jumps, benchmark results and pricing changes across OpenAI, Google, Anthropic, Meta, Mistral and the open-source community.
Days after Kimi K3 launched, Chinese lab Moonshot stopped selling new subscriptions, saying demand had pushed its GPU clusters to the limit — and split its plans into two tiers to ration compute.
Jul 20, 2026
With standard leaderboards saturated and contaminated, Chinese AI leaders argued real-world task performance — and metrics like 'daily active agents' — should replace static benchmarks.
Jul 19, 2026
At WAIC, executives signaled a shift away from uniform low prices toward segmentation, splitting the market into cheap mass tiers, pricier premium plans and free quotas for light users.
Jul 19, 2026
Unveiled at WAIC in Shanghai, Qwen3.8-Max-Preview is Alibaba's first trillion-plus multimodal model — but it launched as a closed API preview, with open weights only promised and the ranking based on Alibaba's own tests.
Jul 19, 2026
Elon Musk says the successor to the 1.5T model behind Grok 4.5 is 'better in every way,' could 'exceed Kimi,' and should keep speed and token efficiency close to today's model.
Jul 19, 2026
From July 20, base usage limits drop about 33% across plans, Fable 5 is confined to Max and Team Premium at half those limits, and Pro subscribers get a one-time $100 credit before paying API rates.
Jul 19, 2026
The Apache-2.0 model reads brain recordings from half a second to 30 seconds long, cleans them up and reconstructs missing channels across setups from consumer headbands to research rigs.
Jul 18, 2026
Five founders, several from Formula 1, are strapping cameras to engineers doing chores and factory tasks; ten months in, Hummingbird leads what the Munich startup calls a record German seed round.
Jul 17, 2026
The Alibaba-backed Beijing lab says K3 is the largest open-source model ever built, and independent benchmarks place it behind only Claude Fable 5 Max and GPT-5.6 Sol Max — with the weights promised for July 27.
Jul 17, 2026
The lab's first in-house model is a sparse mixture-of-experts system trained on 45 trillion tokens and released with full weights — and the company openly concedes it is not the strongest model available, open or closed.
Jul 16, 2026
The London startup's Orbital model fuses time-series data, physics equations and language into one system that can flag an anomaly, chase its root cause and simulate whether the fix breaks something else — with engineering giant KBR leading the round.
Jul 16, 2026
The Alipay parent's robotics unit says it trained a video-action foundation model from scratch for physical manipulation — reporting 93.6% on a 50-task benchmark and a sixfold latency cut to real-robot speeds.
Jul 11, 2026
The first model sold through Meta's new Model API is an agentic coder priced far below OpenAI and Anthropic — and Zuckerberg returned to X after three years to claim Meta's models finally beat Google's.
Jul 10, 2026
According to reports, DeepMind threw out the Gemini 2.5 Pro-based foundation days before launch and started a fresh pre-training run — leaving Google the only major lab without a public flagship as GPT-5.6 and Grok 4.5 ship.
Jul 9, 2026
Elon Musk says the new model rivals Anthropic's Claude Opus 4.7 while running faster and cheaper, priced at $2 per million input tokens — a public launch timed one day before OpenAI's GPT-5.6 reaches everyone.
Jul 9, 2026
The three tiers are reaching ChatGPT, the API and Codex today, ending a nearly two-week government-coordinated preview that had held the model to about 20 vetted partners — though the rollout is staged, not flipped on for everyone at once.
Jul 9, 2026
Hunyuan 3.0 is a 295-billion-parameter mixture-of-experts model that fires just 21B parameters per token, ships under Apache 2.0, and edges GPT-5.5 on science — free on OpenRouter through July 21.
Jul 8, 2026
The rebrand makes it official: the Grok maker no longer exists as a separate company. Its new name signals the plan Musk has pushed since the $1.25 trillion merger — AI compute in orbit.
Jul 8, 2026
The Commerce Department's AI standards body signed off after weeks of testing — ending a preview limited to about 20 vetted partners and making GPT-5.6 the first frontier model to ship on the government's schedule, not OpenAI's.
Jul 8, 2026
General Atlantic is leading the first outside money into China's video-generation champion as Kuaishou preps a Hong Kong spin-off — while Beijing tells its AI firms to think twice about US capital.
Jul 7, 2026
A year ago Chinese models carried under 2% of OpenRouter traffic. Now four of the top five models are Chinese, Xiaomi's MiMo-V2-Pro processes more than 4 trillion tokens a week — and price is the weapon.
Jul 7, 2026
At a July 2 town hall, Mark Zuckerberg said agent progress hadn't accelerated in four months — minutes before AI chief Alexandr Wang claimed a new model caught GPT-5.5.
Jul 6, 2026
The video-AI startup drew backing from NEA, Naver and Amazon to push its Marengo and Pegasus models from search toward reasoning over the world's video archives.
Jul 5, 2026
The new Sol, Terra and Luna models launched to just 20 vetted partners at the US government's request, under a new process for vetting frontier AI before wide release.
Jul 4, 2026
The delayed flagship promises a 2-million-token context window and a Deep Think mode, but Google is still holding it back over token-efficiency and long-task concerns.
Jul 4, 2026
The mid-tier model matches larger systems on agentic coding and computer use, and ships as the default on Claude.ai — but early users flagged tokenizer and API changes.
Jul 4, 2026
