Infrastructure / Deployment

OpenAI Code Interpreter

Closed-source code-execution tool exposed via the Responses API and Assistants API, runs Python (and bash in newer versions) in a fully sandboxed VM with attached files, so the model can compute, analyze data, and iterate on solutions. Picked when teams want a turnkey 'agent can run code' primitive bundled directly with the OpenAI model API rather than wiring up E2B/Modal themselves.

OpenAI Code Interpreter, a tool (Assistants API, migrating to Responses API) that writes and runs Python in a proprietary OpenAI-managed sandbox; processes files up to 512MB, sessions billed at $0.03/session (1-hour default). Confirmed live June 2026 via OpenAI developer docs. Proprietary; isolation mechanism undisclosed.

Openness

1 high confidence

1.0

license: Proprietary
service: proprietary(OpenAI-managed sandbox)
source: closed
isolation: undisclosed
access: API/tool only(no self-host, no open runtime)

Fully proprietary managed code-execution sandbox; OpenAI does not document the isolation mechanism or release any runtime.

https://developers.openai.com/api/docs/assistants/tools/code-interpreter recorded 2026-06-04
Code Interpreter runs Python in a proprietary OpenAI-managed sandbox; $0.03/session; isolation details not disclosed

Adoption

4 low confidence

4.0

Code Interpreter ships inside ChatGPT (advanced data analysis) and the Assistants/Responses API; the ChatGPT surface it rides has ~1B weekly users (flagship anchor record, GPT-5). No standalone Code-Interpreter usage figure is disclosed, so adoption is attributed to the surface, placed conservatively at 4 (a fraction of ChatGPT users invoke code execution) with low confidence given no primary per-tool number.

https://developers.openai.com/api/docs/assistants/tools/code-interpreter recorded 2026-06-04
Code Interpreter is a first-class tool in the Assistants/Responses API and underlies ChatGPT data analysis (the surface that drives usage)

Capability

3 low confidence

3.0

Reliable Python-only data-analysis sandbox with iterative retry, but single-language, undisclosed isolation, and session-scoped, narrower than general-purpose deployment sandboxes. Mid-tier; capability detail limited because OpenAI does not document the isolation/scale internals.

https://developers.openai.com/api/docs/assistants/tools/code-interpreter recorded 2026-06-04
Python sandbox, iterative code retry, 512MB files, 1-hour session

Unchanged since 2026-07-30 (last edited, not re-checked)