OpenAI switches container sessions to per-minute billing with a 5-minute minimum, cutting costs for short runs

OpenAI updated its container session billing on June 2, 2026, moving from a flat 20-minute session rate to per-minute billing with a 5-minute minimum. The change reduces costs for developers whose container sessions complete in well under 20 minutes — the common case for many code execution and tool-use workflows.

What's new

The exact change, per the OpenAI developer changelog: "Starting June 2, 2026, eligible container sessions will be billed per minute with a 5-minute minimum, instead of being billed at the full 20-minute session rate."

The per-minute rate itself is unchanged — developers who consistently run full 20-minute sessions see no difference. The material saving applies to shorter sessions: a 6-minute container run previously cost the equivalent of 20 minutes; it now costs 6 minutes, a 70% reduction for that session. The 5-minute minimum ensures a price floor, so very short sessions of 1-2 minutes are billed as 5 minutes rather than 1.

OpenAI's container sessions power the Code Interpreter and file-analysis tools available in ChatGPT and via the API. They spin up isolated Linux environments where models can execute code, process files, and perform multi-step computations.

Context

The prior billing model was a known source of developer complaint: many container sessions — especially those handling simple computations or quick data processing — resolved far faster than the 20-minute allotment but were billed at the full rate regardless.

This change arrives shortly after OpenAI announced several other developer-facing improvements: workload identity federation for keyless API authentication (May 26), extended prompt cache retention defaulting to 24 hours (May 29), and inline moderation scores in the Responses and Chat Completions APIs (June 4).

Why it matters

For developers running container sessions at scale — particularly those processing many short documents or running quick code-verification tasks — the change meaningfully reduces costs. A workload that spins up 1,000 container sessions per day, each completing in 7 minutes, was previously billed for 20,000 minutes; it is now billed for 7,000 minutes, a 65% cost reduction for that workload.

The move also signals that OpenAI is paying attention to the granularity of its developer pricing. Per-minute billing with a floor is a standard practice in cloud compute pricing — aligning OpenAI's model more closely with the billing norms developers encounter in the rest of their infrastructure.