Welcome to Prompt & Circumstance – your weekly deep dive into the evolving world of Generative AI and its impact on marketing and communications.    

Like this newsletter? Share this with a friend or colleague! 

OpenAI’s Heck of a Week: ChatGPT-5, $1 Government Deals, New Open-Source Models, & a $500B Valuation

A lot happened this week at OpenAI. While ChatGPT-5 may be the talk of the town (and rightfully so), other OpenAI updates flew slightly under the radar. From $1 government deals to the release of an open-source reasoning model and a swift reversal on search privacy.  

Here’s everything you need to know to stay caught up with OpenAI’s busy week.  

(Cue Drum Roll)... The Latest Updates on ChatGPT-5, OpenAI’s Newest Model

Yesterday, OpenAI officially launched GPT-5, the most powerful and human-like version of ChatGPT to date. If you’ve logged into ChatGPT today, you may have already been upgraded, if not, you should get access very soon! 

Unlike previous releases, GPT-5 is more about user experience than about significant upgrades to any one model. Previously, the user had to choose between different types of models depending on the complexity of the task. As a result, according to OpenAI, most people have never used one of their reasoning models and were not clear that model selection was possible on the platform. With GPT-5, OpenAI is removing this friction with an orchestration layer that routes your queries based on complexity. Simple questions get answered quickly with lightweight models; more strategic or nuanced prompts are escalated to a deeper reasoning engine. This makes ChatGPT far more useful for the layered demands of marketing and comms.  

While most of the focus of GPT-5 is more about user experience, testing has shown improvements in the underlying models. Based on LLM Arena’s leaderboard, GPT-5 is the most sophisticated model on the market for both text, web development, and vision, trailed only slightly by Gemini 2.5 Pro. GPT-5 achieved a benchmark score of 1481 based on testing. Its previous o3 reasoning model rated a 1450, and still ranks 5^th best. So it's an improvement, but not a significant leap forward. 

What Makes GPT-5 Better?

• Improved writing capabilities: The model now understands nuance, rhythm, and structure better, especially in formats like headlines, memos, or campaign copy. It can interpret tone more subtly and adapt writing styles with greater flexibility. For comms and marketing teams, this makes it a more reliable writing partner. 

• Fewer hallucinations (i.e. made-up facts): GPT-5 significantly reduces factual errors compared to earlier models. For marketing and comms teams relying on AI to generate public-facing content, fewer hallucinations means greater trust in the outputs, especially when accuracy and brand credibility are on the line. 

• Available with Microsoft’s Copilot: As a way to position GPT-5 as the intelligence layer within enterprise workflows, Microsoft made GPT-5 available in Microsoft 365 Copilot and Copilot Studio on day one. Some users may gain access there even before it's fully live in ChatGPT.  

• Advanced coding: GPT-5’s demonstrated some significant improvements in coding capabilities over earlier OpenAI models. GPT-5 is now more capable of building and debugging websites, apps, and tools from a prompt. This matters for teams experimenting with interactive content, microsites, or internal dashboards. 

• Better safety and contextual awareness: GPT-5 is more cautious, context-sensitive, and reliable when handling complex or sensitive questions. This makes it more dependable for generating messaging, employee guidance, or campaign content that touches on nuanced topics, without crossing a line. 

• Voice mode upgrades: Personalized, adaptive conversations are expanding to more users. This means easier voice-driven brainstorming, quicker content reviews, and more natural interactions when multitasking or working remotely. 

• Preset personalities: New tones like Cynic, Listener, and Nerd can be selected to shape how ChatGPT interacts with you through Custom Instructions. These let marketing leads test copy through different audience lenses or simulate feedback from various personas before publishing.  

Despite all the upgrades, GPT-5 still has limits. 

It doesn’t learn from your behavior or evolve over time on its own, meaning it won’t improve based on your inputs unless retrained by OpenAI. This lack of continuous learning is one of the key reasons it’s not yet considered super intelligence or artificial general intelligence (AGI). 

It’s also not immune to mistakes. GPT-5 can still hallucinate or offer vague or overly agreeable answers in sensitive contexts. While it's more helpful than ever, it's still a tool that needs to be applied with purpose. 

What else happened this week in the Open AI world ….. lots! 

OpenAI Removes Search Indexing of Shared Chats After Backlash 

What started as an effort to showcase helpful AI conversations was quickly rolled back by OpenAI this week following user outcry.   

• A Fast Company report found that nearly 4,500 shared ChatGPT conversations were being indexed in Google search with some containing deeply personal or sensitive info. One chat described someone’s mental health struggles; another included family details and location.  

• These chats were only indexable if users opted in by checking a "Make Discoverable" box when sharing a link. 

• After public backlash, OpenAI removed the feature and de-indexed the content. 

Why It Matters: OpenAI has struggled with perceptions that it is not safe or secure as a platform. This week’s issues with Google search reinforced that view with lots of people reading headlines, rather than understanding the sharing of chats had to be deliberate. This won’t help ChatGPT with enterprise customers in the short term.   

 ChatGPT Enterprise is Now $1 for U.S. Federal Agencies 

• OpenAI announced a new partnership with the U.S. General Services Administration (GSA) to offer ChatGPT Enterprise to all federal agencies for just $1 per agency for the next year. Sorry rest of the world :( 

• The offering includes unlimited access to OpenAI’s most advanced models and features, including Deep Research and Advanced Voice Mode, for a 60-day introductory period. 

• Participating agencies also get onboarding support, including government-specific training resources, via OpenAI Academy and partners like BCG and Slalom. 

• The plan is designed to cut bureaucratic overhead, improve efficiency, and embed AI more deeply into public sector workflows. 

Why It Matters: This ultra-low pricing strategy (watch out Dollar Tree) is a market share play by OpenAI. It gives the company a powerful foothold inside the federal government just as agencies ramp up AI integration. This is also a threat to Microsoft’s dominance with enterprise and government AI. How will both Microsoft and Google respond and what will governments in other parts of the world expect as a result as it relates to pricing?  And how long before the bait and switch? Compute costs are still very high and while OpenAI can raise money easily enough, it won’t last forever.   

OpenAI Releases Two Open-Weight Reasoning Models 

Back in March, OpenAI signaled that open-weight models were coming, especially after DeepSeek shook up the ecosystem with its high-performing, open alternative. Now, they’ve delivered on that promise. 

• The release marks OpenAI’s first open-weight models since GPT-2, a major shift from its closed-source history. 

• The two models, gpt-oss-120b and gpt-oss-20b, are downloadable, customizable, and can run entirely offline, which are ideal for security-conscious organizations. 

• They’re designed with chain-of-thought reasoning that helps the model break down complex tasks step-by-step. 

• Teams can fine-tune these models for proprietary workflows or behind-firewall deployments. 

• Both models are available now on Hugging Face under an Apache 2.0 license. 

Why It matters: This is a tactical counter to Meta's Llama and China's DeepSeek models, and a well-timed release. The two models rank slightly below ChatGPT’s o3 reasoning model. With the launch of GPT-5, there’s less of a reason to keep these models closed source. For marketing and comms, it expands the possibilities for building and fine-tuning models for internal use that operate securely behind firewalls, with no external data exposure. Chat with us if you want to learn more!  

OpenAI Eyes $500B Valuation via Secondary Share Sale 

As OpenAI pushes technical boundaries, it's also expanding its financial strength. This week, news broke of a major valuation milestone that could reshape how the market views AI's long-term potential. 

• OpenAI is reportedly exploring a secondary stock sale that could value the company at $500 billion, up from $300B in March 2025. 

• The move would allow current and former employees to sell shares, providing liquidity amid rising investor demand. 

• CEO Sam Altman is also reportedly leading a $40 billion fundraising effort to retain top talent and align equity incentives. 

• Analysts see this valuation surge as a reflection of ChatGPT’s commercial traction and OpenAI’s dominant position in enterprise AI. 

Why It matters: OpenAI’s aggressive expansion and growing valuation show investor confidence in its business model, but also raise expectations. As OpenAI positions itself as the backbone of enterprise AI, the pressure will increase for them to accelerat adoption of their tools. 

That’s All the OpenAI News for Now! 

It is clear that over the past few months, OpenAI has expanded at a remarkable pace. If this trajectory holds, GPT-5 may just be the midpoint of an even more transformative year ahead. Wouldn't be surprised if ChatGPT-10 is closer than we think. 

Ready to put Gen AI to work? Book a 30-minute call with us and learn how we help teams move beyond basic Gen AI application and into high-impact execution.