Who shipped what across the AI industry, read from official company blogs and changelogs. Read the method.
| Day | Launches |
|---|---|
| AUG 15 | 1 |
| AUG 16 | 0 |
| AUG 17 | 2 |
| AUG 18 | 3 |
| AUG 19 | 2 |
| AUG 20 | 2 |
| AUG 21 | 2 |
| AUG 22 | 0 |
| AUG 23 | 0 |
| AUG 24 | 4 |
| AUG 25 | 3 |
| AUG 26 | 2 |
| AUG 27 | 3 |
| AUG 28 | 1 |
| AUG 29 | 1 |
| AUG 30 | 0 |
| AUG 31 | 1 |
| SEP 01 | 7 |
| SEP 02 | 1 |
| SEP 03 | 6 |
| SEP 04 | 0 |
| SEP 05 | 0 |
| SEP 06 | 0 |
| SEP 07 | 0 |
| SEP 08 | 2 |
| SEP 09 | 1 |
| SEP 10 | 3 |
| SEP 11 | 2 |
| SEP 12 | 0 |
| SEP 13 | 0 |
The 12 busiest companies across the last 5 ISO weeks. One colour step per launch, so a cell is a lookup, not a guess. Empty cells shipped nothing.
| Company | Week of AUG 10 | Week of AUG 17 | Week of AUG 24 | Week of AUG 31 | Week of SEP 07 |
|---|---|---|---|---|---|
| OpenAI | 0 | 1 | 3 | 4 | 2 |
| xAI | 2 | 1 | 1 | 3 | 0 |
| Anthropic | 1 | 0 | 3 | 1 | 0 |
| 0 | 0 | 1 | 4 | 0 | |
| Adobe | 0 | 1 | 0 | 1 | 1 |
| Cohere | 0 | 0 | 3 | 0 | 0 |
| Databricks | 0 | 0 | 0 | 0 | 3 |
| Mistral | 0 | 1 | 1 | 0 | 1 |
| Amazon AWS | 0 | 1 | 1 | 0 | 0 |
| Stability AI | 0 | 1 | 1 | 0 | 0 |
| Apple | 0 | 0 | 0 | 1 | 0 |
| DeepSeek | 0 | 0 | 0 | 0 | 1 |
One row per company that shipped. The trace is that company's own daily launches across the same 30 days, on one shared vertical scale, peak 3 in a day; the bar is that same 30 day total. STREAK marks a company that shipped in every one of the last 3 or more ISO weeks. The 7 that shipped nothing are on the Silence Rail below.
What kind of work is shipping. One bar, the 49 launches inside the same 30 day window as the registers above, split by category. The extractor picks one label per launch from a fixed vocabulary of 13; every label that occurred is defined under the bar. Shading is share, largest first, on the same four step ramp the heatmap uses; the bar and the key run in the same order.
Every tracked launch on the one date spine, newest first. Filter by company, by category, or search the text.
ChatGPT Enterprise workspace owners and admins can now designate a workspace member as a group manager and delegate selected group administration, such as analytics, usage limits and requests, or permitted feature and model permissions on assigned roles. The designation is separate from the member's built-in workspace role.
OpenAI added Box, Dropbox, and SharePoint to the ChatGPT Library alongside Google Drive. Once connected, users can browse and search those files, add them to a conversation through Add from Library or at-mentions, keep them open beside the chat, and follow citations back to the source, on the web for Go, Plus, Pro, Business, Edu, Healthcare and Enterprise, with mobile support to follow.
DeepSeek released DeepSeek-V4.1-Flash, a 552B-parameter mixture-of-experts model using a causal encoder-decoder architecture that activates 8B parameters for input and 16B for output. It is live on the DeepSeek API as deepseek-flash with native multimodal support, and its KV cache needs a quarter of the HBM and an eighth of the SSD storage of the previous generation.
Mistral helped a European energy operator migrate a 40,000-line Fortran 77 reservoir simulator to modern C++ using AI agents. The team built a numerical parity harness first, spawned over a hundred agents to document the codebase, and settled on a human-supervised coder, tester and reviewer workflow after full agent autonomy produced Fortran retyped in C++ syntax.
Adobe unveils AI-powered innovations in Premiere and After Effects aimed at enhancing speed, flexibility, and control for video editing. Key features include the Generative Media Tool, smarter audio tools, and an AI Assistant to automate tedious tasks.
XAI published the design story behind Grok Bot, an interface built around persistent agents instead of chat sessions. Bots get their own name, avatar, memory, tools and cloud computer, the avatar's motion carries the work state, and Routines let a Bot start work on a schedule or an event without a prompt.
Grok Bot is now available for enterprise customers, offering free usage for two weeks. This release enhances enterprise capabilities with access, network, and audit controls, enabling organizations to manage bots effectively at scale.
OpenAI released GPT-6 Astra in the API, describing it as its most capable model, built for the hardest end to end work across reasoning, coding, computer use, research, and document creation. The changelog notes constraints including no none reasoning effort level, no custom temperature, top_p or log probabilities, and tool calling through the Responses API rather than Chat Completions.
OpenAI introduced GPT-6 Astra in ChatGPT with improved coding, research, computer use, and complex multi-step workflows. Initial access is limited to select organizations, with broader availability planned over the following days.
Google DeepMind and Google Research introduced WeatherNext 3, a global weather model that ingests live geostationary satellite data and produces hourly forecasts at 5 kilometer resolution, up from 25 kilometers. Google reports up to a 60 percent improvement in medium range precipitation accuracy and is rolling the model into Search, Gemini, Maps and Cloud.
Google introduced Gemini 3.8 Flash, its most intelligent workhorse model, at the same introductory price as 3.7 Flash of 0.75 dollars per million input tokens and 3.75 dollars per million output tokens, with gains in software engineering, agentic tasks and multi step reasoning. It also introduced Gemini 3.8 Flash Cyber, a cybersecurity model for vulnerability detection and automated patching, available to trusted defenders through the new Fairwind Program.
LatchBio published an independent analysis of Grok 4.6's biosecurity behaviour. The evaluation found the model refused hazardous biology requests while still completing routine biological work.
Perplexity introduced hybrid compute on Mac, letting Perplexity Computer split each task between cloud models and a compact local model on an Apple silicon Mac, with an on device privacy gate that can mask, keep local, refuse or ask for consent before protected information leaves the machine. It is available to Pro, Max and Enterprise subscribers with three local models, Gemma 4 E4B, Qwen3.6 35B-A3B and a Perplexity model, on any Apple silicon Mac running macOS 15 or later with at least 24GB of unified memory.
OpenAI set out its assessment of the Astra model's cybersecurity capabilities and the safeguards applied before release. It says Astra meets a high capability threshold that calls for enhanced protection against misuse, and describes the evaluations and alignment measures used to deploy it safely.
Google launched agentic video understanding across Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite, letting the model navigate a video timeline with native video tools instead of decoding at a fixed frame rate. Google says it cuts token consumption by up to 88 percent and costs by up to 66 percent while improving quality by up to 7 percent, and it is available today via the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform at standard token pricing.
In August 2026, Google announced several key AI updates, including the launch of Gemini 3.7 Flash and the Pixel 11 series of devices. The updates aim to enhance AI accessibility and practicality for various users, including students and professionals, with new tools and hardware.
Apple presents REFACTOR-VLA, a system designed to enable the learning of reusable skills in vision-language-action models through a unique "wake/sleep" architecture. The study highlights key findings related to the scaling of world models and the importance of auxiliary training objectives for skill clustering performance.
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, its most capable models for coding and knowledge work. Both add stronger research capabilities.
Grok Bot has introduced a tighter integration with X, allowing users to connect their X accounts and receive developer accounts if needed. Paid users also benefit from free X API credits.
Cohere published a guide to using generative AI in business, covering content creation, conversational assistance, and data analysis. It sets out the productivity and personalisation benefits alongside the challenges organisations hit when adopting it.
Cohere argues that forward-deployed engineers should build customer capability rather than dependency. The piece compares deployment strategies and describes how FDEs help enterprises reach operational independence while integrating AI systems.
The Model Hardware Standard (MHS) is introduced as a shared specification enabling AI agents to operate physical lab devices safely and efficiently, reducing setup time for experiments. MHS allows operation of multiple instruments in parallel by integrating AI into hardware control.
Claude is enhancing support for the scientific community by launching a new program that provides 10,000 free and discounted subscription seats for researchers. This initiative is part of a broader strategy to help scientists access computing resources and tools tailored for high-impact projects across multiple scientific fields.
OpenAI expanded ChatGPT for Teachers to 55 additional school systems across 20 states, reaching over 100,000 more educators and staff. It now works with more than 100 K-12 organizations across 30 states, covering over 300,000 educators.
Google describes running long-context multimodal embedding inference on Cloud TPU.
Stability AI announced a Series B fundraise of $76M, bringing total funding to $232M. The new round includes investors from leading companies in the entertainment sector such as Electronic Arts and Sony Music Group, aimed at enhancing their creative AI products.
Cohere published IDC research surveying over 500 senior decision makers at large enterprises in Canada, the US, UK and Germany. Only 13 percent reported being very widely aware of sovereign AI, and one in three leaders struggled to define it.
Anthropic is launching a $5 million grant program to fund independent research on how AI affects user wellbeing. The initiative aims to build open-source evaluations, enhance standards for AI interaction, and involve experts in the development of these evaluations.
The `codex mcp-server` command is now deprecated, with users advised to utilize the Codex app server instead. For integration with Claude Code, the Codex plugin should be employed.
Mistral and HUMAIN have announced a strategic collaboration to enhance sovereign AI capabilities in Saudi Arabia and the region. This partnership aims to develop localized AI models, particularly in cybersecurity and voice, while addressing the growing demand for AI systems that maintain control over data and operations.
AWS published a walkthrough for connecting Outlook to Amazon Quick for AI-driven email automation.
Grok 4.6 is now available on the Gemini Enterprise Agent Platform, offering features for long-running agents with a 500k context window and configurable reasoning efforts. This latest flagship model is designed for ambitious interactive and visual work, and can be accessed via Model Garden.
Together AI benchmarked GLM-5.3 against GPT-5.6 Sol on DeepSWE, comparing cost, accuracy and reliability. GPT-5.6 Sol led on single-shot attempts while GLM-5.3 came out ahead on value across multiple attempts, which Together AI presents as the better fit for high-volume coding.
OpenAI published an essay on what it calls the Intelligence Age and what it means for the next economy. It argues AI will reshape sectors and expand what people are able to do.
Mistral's Agentic Search enhances AI-driven search results by improving accuracy while reducing latency and token usage. The system integrates a multi-step retrieval process, allowing for effective navigation and verification within complex documents and data, ultimately making enterprise data more accessible through the Mistral Search Toolkit.
Samsung Electronics is set to unveil the Galaxy S26 series on August 27, featuring camera and AI innovations that enhance daily connectivity for users. The event will be streamed live, allowing global access to explore new Galaxy S26 experiences and the latest One UI.
NVIDIA released SkillEvaluator, an open-source tool that measures how verified skills change AI agent performance. The post reports evaluation results gathered across more than 30 NVIDIA products.
Stability AI opened two new ways to use Stable Audio 3.0 in early beta: a DAW plugin that generates audio inside an existing session, and a more advanced generation and editing experience on StableAudio.com. Both run on commercially safe models and users keep full ownership of their outputs.
The Agentic Data Operations Platform (ADOP) streamlines data engineering tasks, allowing teams to transition from spending weeks on tasks like ETL to focusing on delivering data products rapidly. Designed for teams involved in data governance and compliance, ADOP automates the data lifecycle with AI agents while ensuring consistent architecture and regulatory adherence.
Groq LLC has announced a $350 million Series A fundraise, valuing the company at $3.5 billion. The funding will support Groq's expansion as it continues its partnership with NVIDIA, to enhance AI inference technology and scale its operations.
ElevenLabs has introduced voicemail detection for Conversational AI, enabling voice agents to identify voicemail systems and conserve credits. This functionality allows agents to leave targeted messages or end calls immediately upon detecting a voicemail, improving efficiency during outbound campaigns.
Grok Bot changes product management by allowing PMs to effectively manage a team of AI agents that assist with various tasks like shipping software and prioritizing work.
Grok 4.6, the latest coding model, is now available in GitHub Copilot. Users can start using it by selecting 'Grok 4.6' in the model picker, and more information is available through the SpaceXAI console.
Future Claude models will generate text containing a watermark to determine the likelihood of Claude's involvement in writing the text, in compliance with the EU AI Act. The watermarking method does not affect the quality or content of the outputs, as it remains undetectable to readers and does not carry identifying information.
Past the last day node the spine keeps going and nothing hangs off it. 7 of the 25 tracked companies published nothing at all in the window. Silence is data, so it gets the same mark with a flat path.
The company with the most launches in the current ISO week, from SEP 07. A tie is broken by the most recent launch and every tied company is named. The 3 entries below are Databricks's own, newest first.
3 launches, the most of any company in this window.
Entries come from official company blogs and changelogs only, no aggregators, no social posts. Titles are never rewritten: they are the source's own headline. Summaries are machine extracted, then passed through a deterministic scrub applied at render.
Every launch keeps its source link, so every row is traceable. Link provenance is the one state this page can compute, and it is recomputed on every build: 51 of 51 indexed launches carry a first party URL, across 24 distinct hosts. Nothing else here is independently verified. It is a scrape.
Date sanity: any post that parses to a future date is treated as today, so a future date never appears on this page.
The scrub is rules, not a model, so the same input always produces the same output and data.json is never hand edited: 51 summaries checked this build, 13 rewritten. It removes seven things and nothing else: abstractor openers, whole abstractor sentences, participial abstractor tails, vendor first person, vendor superlative adjectives and adverbs, hype verbs, and participial hype tails. A tail that runs more than half again as long as the clause it hangs off is kept, because at that length it is the content, not an aside.
It never rewrites a finite verb the source chose. 11 of the 44 printed summaries therefore still say "enhance" somewhere, and 3 still say "aims to": those are the source's own claim and its own stated intention, and upgrading an intention into a shipped fact would be a worse defect than the hedge it removed.
An abstractor opener is removed outright when what follows can stand as a sentence (0 this build) and promoted into the claim hiding inside it when there is one (0, a reduced relative clause or an appositive definition). When the source sentence carried nothing but what the post is about, the abstractor's subject is replaced by the publisher, which is true and gives the sentence a subject and a finite verb, and that is what the page prints (2). No strip may leave a fragment: 0 were reverted for that reason.
Feed weight is computed from the entry, not chosen by hand: 20 lead, 24 standard, 7 rolling document updates.
Data scraped SEP 13, 2026. Pulse refreshes weekly.
Grok Bot is revolutionizing product management by allowing PMs to effectively manage a team of AI agents that assist with various tasks like shipping software and prioritizing work. This blog discusses how Grok Bot integrates with PM workflows, providing use cases and insights into its impact on productivity and collaboration.
Grok Bot changes product management by allowing PMs to effectively manage a team of AI agents that assist with various tasks like shipping software and prioritizing work.
171 editions so far, one a day. Unsubscribe anytime.