OpenAI GPT-5.4 aims to close Anthropic enterprise gap

OpenAI's launch of GPT-5.4, a model designed for professional work, on the surface is about closing the buzz gap with Anthropic, which has captured spending with Claude models and Claude Cowork, but the bigger picture is that the LLM race is really about creating productive systems.

In its launch post, OpenAI said GPT-5.4 Pro in ChatGPT and the company's API brings together our "our recent advances in reasoning, coding, and agentic workflows into a single frontier model." GPT-5.4 also includes the coding capabilities of GPT-5.3-Codex and now works better across tools, software and tasks with spreadsheets, presentations and documents. For good measure, GPT-5.4 has integrations with Microsoft Excel, Google Sheets and financial services data and tools.

OpenAI is building on its recent enterprise launches and Frontier.

• OpenAI’s 2026 focus on practical AI points to enterprise
• Amazon, OpenAI forge multi-faceted partnership: Dissecting the deal
• OpenAI launches OpenAI Frontier, aims to be AI agent orchestrator

Yes, this GPT-5.4 launch sounds a lot like an Anthropic Claude Cowork or Claude Code launch.

• Anthropic vs. SaaS: A nuanced view
• Anthropic expands Cowork plugins across enterprise functions

What will ultimately make GPT-5.4 successful is its ability to grab context from other systems, get work done quickly and efficiently and orchestrate well with its various OpenAI models. These LLM launches today are more about leveraging multiple models in a multimodal way than any one evaluation or benchmark.

Microsoft CEO Satya Nadell noted as much this week at a Morgan Stanley investor conference. The game isn't at the LLM level but the orchestration and context level. Nadella said:

"If you're building any product, whether it's for coding agents or for knowledge work or wealth management or whatever, you want to access multiple models. That's going to be the case. And that means there's lots of very careful design that you want to do, which is you want to have the harness not get coupled with the model layer. So, people are getting pretty sophisticated in making sure that the harness layer is decoupled.

The other is the context layer also should be decoupled, right? You don't want effectively the one model to, in fact, vertically integrate into these two. And that will be the game that will be played. But I'm pretty clear industry structure-wise, where we will go."

• How "Context Graphs" are Redefining the AI Landscape
• Why the Next AI Wave Is About Process, Not Models, Part 1
• Box CEO Levie: AI agents need context, unstructured data
• Agentic Automation Doesn’t Fail at Reasoning. It Fails at Inputs.
• Welcome to the context chorus: There’s no AI without context

What OpenAI's GPT-5.4 launch and Anthropic's recent Claude efforts represent are vertically integrated model and system launches. Ultimately, OpenAI and Anthropic will want to manage multiple models beyond their own. That's where the enterprise AI game gets really interesting. Nadella's point makes sense: The game is at the context layer, which will need to be its own layer.

What you need to know about GPT-5.4

Here’s what you need to know about GPT-5.4.

• OpenAI GPT-5.4 Thinking is the headliner of the launch because it can provide an upfront plan of thinking and enable you to adjust it mid-response. It'll also have deep web research and maintain context longer.
• GPT-5.4 has native computer use capabilities so agents can operate computers across multiple apps and workflows.
• Agents created by GPT-5.4 are more reliable across various benchmarks compared to its predecessor models.
• In the API, GPT-5.4 includes a tool search to look up a definition and include it in a conversation. The technique lowers the tokens required in a workflow to improve efficiency and lower costs. OpenAI is messaging more to AI value and returns.
• GPT-5.4 Thinking is better at pulling information across multiple sources on the web and identifying the most valuable ones. OpenAI needed to improve on its web search. My queries have indicated Google Gemini easily outperforms GPT-5.2.
• ChatGPT for Excel was launched in beta to build and update spreadsheets. Financial data integrations were also added into ChatGPT.
• And finally, GPT-5.4 Pro isn't cheap. Here's a look at the pricing.