AI Roundup September 2026: Model Updates and Software Launches

There’s always something happening in the AI industry, and September 2026 was no different. The month opened with several model updates, followed by new personal agents and tools for marketing, analytics, customer service, and creative work.

Here’s a quick roundup of the month’s noteworthy launches and updates.

Model updates in September

1. OpenAI

In September, OpenAI introduced GPT-6 Astra, with improvements in computer use, software engineering, scientific reasoning, and creating documents, spreadsheets, and presentations. Its rollout began with selected organizations before expanding to eligible paid ChatGPT users and developers.

On September 22, OpenAI added GPT-6 Sol and Luna as faster, lower-cost options for everyday work. GPT-6.1 Sol followed on September 29, bringing further improvements in coding, computer use, and professional tasks at one-fifth of Astra’s standard API input and output token prices.

OpenAI also introduced Dots on September 29, persistent personal agents that carry out ongoing work using their own cloud computer, browser, and connected apps. Dots can manage multiple tasks, retain preferences, and bring completed work back for review. The rollout began on eligible paid plans in supported markets, alongside a preview of specialist dots for organizational responsibilities.

2. Anthropic

Anthropic opened September with Claude Fable 5.1 and Mythos 5.1, introduced on September 1 with improvements for long-running coding and knowledge work. Fable 5.1 provides broader access with additional safeguards, while Mythos 5.1 remained restricted to vetted organizations for advanced cybersecurity and biology work.

Claude Opus 5.5 arrived on September 22, followed by Sonnet 5.5 on September 28. Anthropic reported more efficient execution, including a 40% reduction in Opus task costs and up to 30% lower costs per task for Sonnet compared with their predecessors.

3. Google

Google’s September model releases began with Gemini 3.8 Flash and Flash Cyber on September 2. Flash introduced improvements for coding and multistep tasks, while Flash Cyber focused on vulnerability detection and patching, with access restricted to trusted defenders.

Gemini 3.8 Live and Live Extended Thinking followed on September 15 for real-time conversations and complex voice-driven tasks. On September 30, Google announced Gemini 4 Argon for demanding professional workflows, with initial access limited to trusted cybersecurity defenders through its Fairwind Program.

Google also launched agentic video understanding on September 1 for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. The feature lets models search and revisit relevant frames, audio, and transcripts instead of processing videos at a fixed sampling rate. Available through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform, it reduced token use by up to 88% and analysis costs by up to 66% in Google’s benchmarks, while improving accuracy by up to 7%.

4. ElevenLabs

On September 28, ElevenLabs released Eleven v4 and Eleven v4 Turbo, introducing more expressive speech, improvements to voice cloning, and more natural conversations involving multiple speakers. Turbo is the lower-latency option for applications that need quick spoken responses, including real-time voice agents.

5. Perplexity

In September, Perplexity expanded Portable Computer to Windows, bringing its local AI agents to compatible NVIDIA RTX systems with at least 24 GB of GPU memory. Released on September 14, the update lets users analyze files and carry out multistep tasks locally, with cloud models available when additional reasoning is needed. Locally completed work does not consume Perplexity Computer credits, and the agent asks permission before sending information off-device.

AI software launches and updates in September

1. Adobe

On September 8, Adobe Premiere gained tools for generating video and sound effects directly in the timeline, alongside beta capabilities for music, soundscapes, and audio cleanup.

On September 9, Acrobat introduced interactive reports, summary slides, audio summaries, and document styling with Adobe Express templates.

On September 24, Adobe began rolling out Photoshop, Lightroom, Express, and Firefly tools inside Gemini. Its Claude plugin also gained Acrobat tools and interactive editing for PDFs and Express designs, allowing users to refine content directly within the conversation.

2. Haus

Haus helps marketing and finance teams measure the effects of advertising and guide spending decisions. On September 21, it integrated Architect, its AI marketing decision agent, into Causal MMM, its marketing mix modeling system.

Existing Causal MMM customers gained access to investment recommendations with supporting evidence and confidence estimates, followed by evaluation of the changes made.

3. Attentive

Attentive helps retail and ecommerce brands manage personalized marketing across SMS, email, and other messaging channels.

On September 22, it added Lifecycle Intelligence to AI Pro, covering product interests, purchasing behavior, and customer lifetime value, with Product Affinity generally available and the other capabilities in beta. A conversational Reporting Agent and connections to external assistants such as Claude and ChatGPT also launched in beta.

4. Databox

Databox combines business performance data with an AI Analyst for answering questions and preparing reports.

Routines launched on September 7, allowing users to schedule a prompt or reusable Skill and receive the results automatically. The feature turns recurring analysis and reporting into a scheduled workflow instead of requiring someone to start each review.

5. Typewise

Typewise provides AI software for managing customer service conversations and workflows. The company announced Nova, an AI Operator that turns plain-language instructions into customer service operations, tests them on real conversations, and adapts them as the business changes. Teams approve what goes live, with Nova designed to make the platform accessible without a dedicated AI specialist.

6. Loqua

Loqua is a voice-powered AI assistant that turns spoken ideas into polished writing and answers questions about content on your screen. Its launch introduced dictation, screen-based questions through Capture to Ask, read-aloud features, and support for around 100 languages on macOS and Windows. Mac users can also perform actions by voice, while longer, multistep tasks remain part of its planned development.

7. Narrative

Narrative is an AI video editing tool that lets creators edit footage, create motion graphics, and refine videos through chat.

Its launch introduced an editing agent that can use reference videos to guide the style, adjust pacing, add animated titles, change music, and apply effects within the same workflow. Rendering and storage are handled within the platform, so creators can continue refining their videos without moving to another editing tool.

8. ZeroClick

ZeroClick helps businesses sell APIs, data, software, and online services through storefronts that AI agents can read and purchase from. It introduced tools for publishing services, prices, and purchase instructions, accepting payments, and tracking visits, purchases, and handoffs to a person.

Editor’s note: Availability, pricing, and access details reflect the announcements covered in this roundup. Some features launched in beta or through limited rollouts, and their status may have changed since then. Check the product’s website for current details.