Linked and quoted from the labs, the press and the people using them. Times in UTC.
The 17 labs we track and the biggest stories beyond them. The rest is in Everything.
Tencent released ContextPilot-8B, an 8B text-generation model built on Qwen3, on Hugging Face. The model targets context management, tool use, and agentic conversational tasks.
Cohere introduced Parse, a product for enterprise document intelligence at scale. The announcement was published on Cohere's official site.
Google Research introduced GlucoFM, a foundation model for continuous glucose monitoring. The work is presented under Google's Health and Bioscience research area.
Google released Gemini 3.5 Transcribe, a speech-to-text transcription model. The announcement says it provides more intelligent transcription.
IBM released its Granite 4.2 language models in 3B, 8B, and 30B sizes, trained on about 15 trillion tokens with a context window of up to 512,000 tokens. The larger models use agentic RL training to learn tool use and code execution, and all models are available under the Apache 2.0 license.
Google Research published AgentHands, a method for generating interactive hand gestures for spatially grounded agent conversations in extended reality. The work falls under human-computer interaction and visualization.
OpenAI's Jalapeño chip was tested on SemiAnalysis' InferenceX benchmark. It delivered more tokens per user and more throughput per kilowatt than the current state of the art.
Zai released the GLM-5.3-Flash-BF16 model on Hugging Face. The model is image-text-to-text and supports English, Chinese and Arabic.
Tencent published the WeMM-Embedding-2B model on Hugging Face. It is a multimodal embedding model for text and image embedding, built on qwen3_5.
Google released the TimesFM 3.0 PyTorch model for time-series forecasting on Hugging Face. The model has over 1.1 million downloads and 876 likes.
Anthropic reported an issue logging into Claude.ai. The incident was resolved after about 15 minutes.
Anthropic reported a minor incident causing errors when logging into Claude.ai. The issue was resolved after about three minutes.
Anthropic reported elevated errors on requests to Claude models, including Claude Opus 5 and Fable 5. The incident lasted about 204 minutes and has been resolved.
Alibaba published the Qwen3.8-Flash-Next image-text-to-text model on Hugging Face. The model card lists transformers and safetensors support, conversational use, eval results, and a custom license.
An OpenAI staff member said usage-limit resets were propagated to accounts and fixes were landed for previously reported usage issues. More changes are promised for tomorrow.
Mistral published a company post titled Mistral x HUMAIN dated August 24, 2026. The post indicates a partnership between Mistral and HUMAIN.
Google Research announced an AI tool that prioritizes candidate biomarkers from wearable sensor data. The tool uses generative AI.
Google DeepMind is partnering with game studios to prototype AI gameplay, building on 15 years of AI research in games. The work spans titles from Atari to EVE Online.
OpenAI staff announced a reset of usage limits for Codex and ChatGPT Work. The reset makes limits more generous for users.
Google Research published work on how mobility data can give language models a deeper understanding of place. The item is categorized under Algorithms and Theory.
DeepSeek published API documentation for a vision model called DeepSeek-v4-flash-vision-exp. The model appears to add image understanding to the v4 flash line.
Anthropic is privately hoping its upcoming IPO will match or surpass SpaceX's record-breaking IPO, according to anonymous sources cited by Bloomberg. The company is reportedly pushing to go public ahead of rival OpenAI.
Anthropic reported elevated errors on requests to multiple models. The issue was resolved after about 26 minutes.
Anthropic reported elevated errors affecting Google connectors. The issue has been resolved after about 29 minutes.
Google announced five new ways to use Search tools for studying for classes and standardized tests. The features are aimed at learning and test preparation.
Google released the tipsv1-s14 model on Hugging Face. It is a vision model for zero-shot image classification and feature extraction.
Tencent's Hy-MT2-7B, a 7B-parameter translation model, is now listed on OpenRouter. It supports 33 language pairs plus five Chinese dialect and minority-language pairs, with structured, delimiter-based, contextual, glossary-based, and style-guided workflows.
Anthropic reported degraded performance affecting Claude Opus 5 and Claude Haiku 4.5. The incident, lasting about 80 minutes, has been resolved.
Anthropic extended the 50% higher Claude Code weekly limits through August 31. The company said it hopes to make the increase permanent, though capacity may stay tight.
Anthropic reported degraded performance affecting multiple models, including Claude Opus 5. The issue lasted from 16:11 to 18:23 UTC and has been resolved.
OpenAI is shipping a version of ChatGPT tailored to users aged 13 to 17.
Tencent published the AuK model on Hugging Face for text-to-speech, voice cloning, speech generation, editing and enhancement. The model has 3895 downloads and 353 likes.
Anthropic's annualized revenue has risen to $65 billion. The company added $18 billion in annualized revenue over two months.
Anthropic reported degraded performance affecting Claude Opus 5 and Claude Sonnet 5. The incident, lasting about 93 minutes, has been resolved.
Tencent released the EVIE-Preview-4.5B model on Hugging Face. It is a vision-language model for visual document retrieval built on the colpali-engine.
Google Research published work on estimating cardiometabolic risk using smartphone imagery, moving beyond BMI. The item is a general science research result from the official Google Research source.
Google Gemini and Pixel announced a partnership with five global football clubs. The partnership aims to improve the fan matchday experience using AI and smartphone technology.
An issue affected claude.ai, platform.claude.com, the Claude API, Claude Code, and Claude Cowork. The incident, rated critical, lasted 36 minutes and has been resolved.
Stripe will reportedly acquire AI gateway startup OpenRouter for more than $7 billion. OpenRouter's CEO recently described the startup as Stripe for AI.
Anthropic reported elevated errors affecting Claude Fable 5. The issue was resolved after impacting services from 20:00 Aug 14 to 00:11 Aug 15, 2026 UTC.
Anthropic revealed it has developed an AI model more capable than Claude Mythos 5. The company detailed the algorithm in its latest AI alignment report, which outlines potential risks from its large language models.
An issue affected the Claude API, Claude Code and Claude Cowork. The disruption lasted from 20:14 to 20:38 UTC and has been resolved.
CohereLabs published North-Small-Translate-1.0-fp8, a translation-focused text generation model, on Hugging Face. The model uses the cohere2_moe architecture and is available in fp8 format.
OpenAI launched Ultrafast, an inference mode that runs GPT-5.6 Sol at up to 750 output tokens per second using Cerebras hardware. It joins Standard and Fast to form a three-tier pricing structure.
Meta released Glimmer, an open-weight AI model that anyone can download and run on their own hardware. The release came alongside a letter from Mark Zuckerberg arguing AI should be for everyone rather than controlled by a few labs.
Zhipu AI released GLM-5.3, which it says is the strongest open-weights coding model, with a 50 percent improvement over its predecessor from post-training alone. The model was trained for cybersecurity and helped find 2,436 vulnerabilities across 269 projects; weights are set to go open source in two weeks.
Z.ai published a blog post announcing GLM-5.3, describing it as frontier coding with emergent cyber capabilities.
Tencent released UI-Mate-27B, an image-text-to-text model, on Hugging Face. It is tagged for computer-use-agent and GUI-agent tasks and is built on qwen3_5.
Anthropic reported a service disruption affecting Claude Code on the web and related services. The issue was resolved after impact from 23:42 to 00:50 UTC.
Databricks sought $1B but raised $5B after strong investor demand. The round values the company at $190B.
Google introduced Gemini 3.7 Flash, a new model in its Gemini line.
Google announced Sheets canvas, a feature that turns spreadsheet data into interactive dashboards, study trackers and seating charts from a prompt.
Anthropic reported elevated errors affecting Claude Mythos 5, Claude Fable 5, and Claude Sonnet 5. The issue was resolved after about 95 minutes.
OpenAI staff announced that usage limits for Codex and ChatGPT Work have been reset.
MiniMax announced MiniMax Music 3.0.
Anthropic introduced a watermarking system for Claude outputs. Some users complained on social media that it will reveal their use of Claude at work or in classes.
Amazon will use Twitch streamers' content to train its AI by default unless creators opt out. Twitch CPO Mike Minton said on a livestream that an opt-in approach would result in nobody opting in.
Anthropic reported degraded performance affecting multiple models. The issue has been resolved.
Alibaba published an FP8 quantized version of Qwen3.8-27B on Hugging Face. The model is image-text-to-text, conversational, and Apache-2.0 licensed.
Grok 4.6 from SpaceXAI is listed on OpenRouter with frontier performance on coding, knowledge work, and STEM. It is succeeded by Grok 4.7.