GPT-5.4 launches in ChatGPT as Cursor introduces Automations

OpenAI introduced GPT-5.4 on March 5 and announced its rollout in ChatGPT, the API, and Codex.¹ In ChatGPT, the model is called GPT-5.4 Thinking; GPT-5.4 Pro is also available for particularly complex tasks.¹ The rollout to ChatGPT Plus, Team, and Pro users began that day and is gradual.¹ Developers can access both variants through the API.¹

OpenAI brings reasoning, coding, and tool use for agentic workflows together in GPT-5.4 and positions the model for professional work.¹ For companies, its availability through several channels means they can test tasks in chat, Codex, and their own applications.¹ That alone does not establish whether GPT-5.4 should replace an existing model in a particular workflow. Teams will need to compare quality and costs on their own tasks before switching.

Cursor introduced Automations on March 5, 2026.¹ The feature lets users set up cloud agents that run on a schedule or in response to events in connected services.¹² Users define a trigger, write instructions, and select tools; the agent then carries out the task and checks its output.³ Automations is available at cursor.com/automations.¹

Cursor lists security reviews, incident response, and pull request risk assessments as potential uses.² For engineering teams, the approach makes it possible to tie recurring tasks to events rather than start every agent run manually.² Those workflows are examples: Cursor does not provide measured gains in time saved or output quality in its launch announcements.¹²

Luma introduced Luma Agents on March 5, 2026.¹ The system is designed to work on creative projects across text, images, video, and audio while retaining project context between tasks.¹² Luma says it is powered by a new architecture called Unified Intelligence, with Uni-1 as its first model.¹ The service was accessible through an API at launch, according to TechCrunch.³

Luma positions the agents as an alternative to workflows that require teams to connect several specialized tools themselves.¹ The agents can route tasks to external models and coordinate their outputs within a shared workflow.¹² That could be useful for campaigns spanning multiple media, but it does not establish a performance advantage over existing workflows. The announcements do not provide pricing or verified comparative results for Uni-1.³¹

On March 7, OpenAI presented Symphony as an open-source specification and reference implementation for orchestrating Codex.¹² Symphony takes tasks from project boards and coordinates coding agents as they work on them.³¹ This lets teams connect existing workflows to agent-based coding. Symphony is designed to coordinate Codex rather than serve as a general project-management system.³²

OpenAI says Symphony demonstrates how Codex can work with existing task-management tools.³ The company explicitly does not plan to maintain Symphony as a standalone product.³² For businesses, the published code is therefore better understood as a starting point for their own integrations than as a service with a long-term maintenance commitment.³² The available announcement provides no comparative data on how well this approach works across different development environments.³²

Ai2 released Olmo Hybrid on March 5, 2026. The open 7B model combines transformer attention with linear RNN layers and is available on Hugging Face.¹² Ai2 says it matched Olmo 3’s MMLU accuracy after training on 49% fewer tokens.¹³

That comparison concerns a specific evaluation, not performance across every task. The model card also reports roughly twice the data efficiency on core evaluations during pretraining, along with throughput and memory benefits for long contexts.² The architecture may interest teams weighing the training cost and operation of open models. Those reported benefits still need to be tested against a company’s own tasks, context lengths, and infrastructure before they inform a deployment decision.

OpenAI published a framework with five value models for business AI adoption on March 5, 2026. They cover workforce empowerment, AI-native distribution, expert capability, systems and dependency management, and process re-engineering with agents. The article is intended to help business leaders assess opportunities and set priorities. It is neither a new product nor a calculator for financial returns.¹

OpenAI recommends a phased approach: build fluency and trust first, capture value in a few high-value activities next, then scale deeper changes.¹ It cautions against complex agent workflows before governance and controls are mature.¹ OpenAI also launched a channel focused on enterprise AI adoption that day.² The framework suggests ways to assess business value, but it does not demonstrate financial returns from the five models.¹