OpenAI says GPT-5.6 uses 54% fewer output tokens and 57% less time per task in the Artificial Analysis Coding Agent Index, but the company's new enterprise guidance turns the cost question towards accepted outcomes, admin controls and repeatable workflows.

The official post says token prices fell 97% from GPT-4 to GPT-5.4.

It also says token price alone is not enough to measure AI value, because enterprises still have to track tasks completed, time saved, decisions improved and workflows ready to scale.

GPT-5.6 Metrics Sit Beside Enterprise Spend Controls

The official guidance frames enterprise AI cost around useful work per dollar instead of the lowest price per token.

Cheaper models can still create extra cost when they fail, retry or produce work that needs correction, according to the company.

For priority workflows, customers are told to track cost per accepted outcome.

The examples are a resolved customer-support case or a tested engineering change that passes review.

The company also said teams should evaluate the full cost of reaching a quality bar, including model and tool usage, attempts, completion rate, latency and human review.

Teams can cut repeated model calls by narrowing instructions, limiting tools, reusing context and setting clear stop conditions, the post said.

ChatGPT Work Admin Console Tracks Usage By User, Product And Model

ChatGPT Work supports longer, multi-step tasks, so usage can vary widely by workflow.

The Admin Console gives administrators usage analytics and spend controls covering adoption, credit usage and spend by user, product and model.

Workspace, team, user, product and model views can show whether adoption and spending are moving together, where demand is growing and whether higher-cost models are being used for sustained work.

The controls function as a management layer for ChatGPT Work, not as a claim that every AI workflow has a proven return.

The same guidance separates experimentation from production funding, with representative validation before larger deployments.

Governance Controls Cover Tools, Context And Approvals

The governance section says scaling depends on approved ChatGPT context, available tools, permitted actions, higher-risk approvals and the process for granting extra capacity.

Those settings are tied to wider use of plugins, connectors, Computer Use and frontier tools that can act across enterprise systems.

ChatGPT Work gives administrators controls for access, approved context, connected tools, permitted actions, usage and spend.

Spend-control options cover default workspace settings, group caps, individual exceptions and project-context review requests.

AI Deployment Engineers can work with customers on evals, architecture, latency, reliability and workflow design for priority deployments, according to the company.

Customer Savings And Deployment Counts Remain Outside The Public Record

The investment section describes a portfolio of broad productivity access, function-specific workflows and strategic bets built around proprietary company context.

Production funding should cover integrations, controls, reliability and change management once a workflow passes exploration and validation.

OpenAI Frontier and Deployment Company support enterprise AI coworker projects across internal systems, the post said.

ChatGPT Work covers chat, coding, agentic workflows, connectors, plugins, Computer Use and administration.

Customer-level savings, deployment counts, ChatGPT Work pricing changes and independent third-party validation of the GPT-5.6 efficiency figures remain outside the public record.