Quick answer: In September 2026, OpenAI released GPT-6 Astra, and Google launched Gemini 3.8 Flash (better performance at the same price as its predecessor) and Gemini 3.8 Live (voice models for real-time conversations). For businesses, this means more capable AI agents, better value per task and voice assistants that are closer to production-ready. Key takeawaysGPT-6 Astra launched on 3–4 September, first for a limited set of organisations, with a wider rollout to follow.Gemini 3.8 Flash improved on its predecessor while keeping the same price.Gemini 3.8 Live targets voice agents, supports 97 languages and can carry out tasks during a conversation. September was one of the busiest months of the year for AI model launches. Three releases stand out for businesses: a new flagship model from OpenAI and two families of models from Google. Here is what each one is, and what it means in practice if your business uses, or is planning to use, AI. What launched in September 2026 ModelCompanyDateWhat it is forGemini 3.8 FlashGoogle2 SeptemberFast, lower-cost model with stronger coding, agentic tasks and reasoningGPT-6 AstraOpenAI3–4 SeptemberOpenAI’s most advanced model, first to a limited set of organisationsGemini 3.8 Live and Live Extended ThinkingGoogle15 SeptemberReal-time voice models for building voice agents GPT-6 Astra: OpenAI’s new flagship OpenAI announced GPT-6 Astra in early September and describes it as its most intelligent and aligned model so far. According to reporting on the launch, it was released first to a limited set of organisations, with general availability to follow in the days after. OpenAI highlighted gains in professional work, software engineering, science and cybersecurity, and devoted a large part of the announcement to safety, after a period of growing concern about AI systems acting outside their intended limits. For businesses, the practical point is that each new flagship model tends to handle longer, multi-step tasks more reliably. That matters most for AI agents, which plan steps and use tools such as a CRM, email or calendar to complete work. Gemini 3.8 Flash: better results at the same price Google released Gemini 3.8 Flash on 2 September. Google says it delivers substantial gains over the previous Flash model, particularly in software engineering and complex reasoning, and that it “works harder” on difficult problems by running extra reasoning steps and calling tools repeatedly. $0.75per million input tokens (introductory pricing to 31 December)Source: 9to5Google$3.75per million output tokensSource: 9to5GoogleSameprice as the previous Flash model, with better resultsSource: Enterprise DNA This is the pattern businesses should notice: capability is rising faster than price. Tasks that were too expensive or unreliable to automate a year ago, such as reading and classifying every incoming email or document, are increasingly affordable. Gemini 3.8 Live is built for voice agents that can hold natural conversations and complete tasks during a call. Gemini 3.8 Live: voice agents get closer to production On 15 September Google introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two voice models that Google says “deliver the building blocks for reliable, production-ready voice agents”. The standard model focuses on speed and cost; the Extended Thinking version handles more complex reasoning. 97 languages, with automatic detection and switching mid-conversationTasks in the background while the conversation continues, such as checking availabilityNatural acknowledgements such as “Let me check that…” while it worksAvailable to developers through the Gemini API and Google AI Studio, and in preview for enterprises Google’s examples include multi-step booking and real-time troubleshooting support. For businesses that handle a lot of phone enquiries, this is the area to watch. See our AI voice agent blueprint for how such a system is designed. What this means for your business Test on your own tasksBenchmarks are useful, but the only test that matters is how a model performs on your emails, documents and customer questions.Avoid locking into one modelNew models arrive every few weeks. Build automations so the model can be swapped without rebuilding everything.Measure cost per taskTrack what each automated task costs and how often it needs a person. Prices and quality change quickly.Keep people in controlMore capable agents can take more actions, so keep approvals for anything important and log what the AI does. Good first projects remain the same: instant replies to new enquiries, lead qualification, document processing and answering common questions from your own content. Newer models make these faster and cheaper to run. Frequently asked questions What is GPT-6 Astra? GPT-6 Astra is OpenAI’s most advanced model, announced in September 2026. OpenAI says it improves on professional work, software engineering, science and cybersecurity, and it was released first to a limited set of organisations. How much does Gemini 3.8 Flash cost? At launch, Gemini 3.8 Flash was priced at $0.75 per million input tokens and $3.75 per million output tokens, introductory pricing that runs to 31 December 2026. Check Google’s pricing page for current rates. Should my business switch AI models every time a new one launches? Not automatically. Test new models on your own tasks and switch when the improvement in quality or cost is clear. Designing your automations so models can be swapped makes this easy. We build AI agents, voice assistants and automations that work with the latest models. See our AI automation services or get a free consultation. SourcesAl Jazeera: OpenAI unveils GPT-6 Astra (4 September 2026)9to5Google: Gemini 3.8 Flash launch (2 September 2026)Enterprise DNA: Gemini 3.8 Flash, same price, better benchmarksGoogle: Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking (15 September 2026)