OpenAI Open-Sources Codex Harness Framework

OpenAI released the Codex Harness engine under Apache-2.0, boosting GPT-5.6 Sol scores while slashing token costs sixfold.

@sharedot

OpenAI Open-Sources Codex Harness Framework

By @sharedot · · 7 pages

  • Tech
  • Artificial Intelligence
  • Open Source
  • Openai
  • Codex

OpenAI released the Codex Harness engine under Apache-2.0, boosting GPT-5.6 Sol scores while slashing token costs sixfold.

What Happened

On 20 August 2026, OpenAI officially open-sourced Harness, the core underlying engine and execution framework that powers its flagship coding agent, Codex.

The release was made under the Apache-2.0 licence, allowing developers to freely modify, embed, and commercialise it without being restricted to a general-purpose chat interface.

The release includes three components: the codex exec CLI tool, the official Codex SDK supporting TypeScript and Python, and the app-server, which serves as the core engine execution server.

All three sources in the pack confirm these details, and Open Source For You reports that Harness manages the execution loop for AI agents, including task comprehension, long-conversation memory retention, real-time event streaming, tool invocation, interruptibility, status synchronisation, and human-in-the-loop approval workflows.

Why It Is Surprising

The open-sourcing of a core AI agent engine is notable because it hands developers the same infrastructure OpenAI uses internally, rather than just an API endpoint.

According to 36 Kr, OpenAI President Greg Brockman excitedly reposted on X, stating that "Codex can power far more than programming tools."

The same publisher reports that OpenAI's official blog argued against forcing every team to move their familiar workflows into a generic code assistant, instead advocating bringing agents directly into software designed around actual work.

36 Kr also notes that the well-known developer @ClusterProtocol remarked that separating the front-end approval interface from the underlying execution loop makes embedding agents into business dashboards easier than ever before.

This represents a shift away from the chat-box paradigm that has dominated AI tooling.

The Evidence

OpenAI released benchmark data showing how critical the Harness design is to agent performance. All three pack sources confirm that on the ARC-AGI-3 benchmark, optimising the Harness framework via retained reasoning and context compression improved the GPT-5.6 Sol model's score from 13.3% to 38.3%.

Additionally, all three sources report that this optimisation reduced token consumption by sixfold, or to one-sixth of the original output token volume, significantly lowering API call costs.

Open Source For You reports that enterprise organisations including Cisco, for cloud management tools, and Thrive Holdings, have already deployed custom AI agents built on the Codex Harness framework.

36 Kr adds that the GitHub repository is available at openai/codex for developers to access all three open-source components.

The Stakes

The release has immediate real-world implications across multiple industries. Open Source For You reports that a tax-prep pilot using the framework processed 7,000 returns and reduced overall preparation time by approximately one-third.

36 Kr reports that partners Thrive Holdings and Crete integrated the open-source Harness into their professional tax preparation workflow, handling complex tax logic and incorporating feedback from tax consultants.

The same publisher reports that Cisco used the Codex SDK to build App Builder within its cloud control platform, enabling customers to create custom applications using natural language while the complex underlying logic is handled silently by Codex Harness in the background.

These early deployments demonstrate that the framework extends well beyond coding into vertical professional tools.

What Comes Next

According to 36 Kr, OpenAI officially demonstrated a virtual logistics operation dashboard application named Relay to illustrate the framework's potential.

The publisher describes a workflow where users select a delayed waybill and click a comparison button, the application automatically feeds waybill details and logistics data to the AI as context, and Codex calls the application's own tools to obtain real-time data.

36 Kr reports that after the agent analyses an optimal re-booking solution, an approval box must appear, and AI will not perform writing operations until business staff click "Agree." The Apache-2.0 licence means anyone can modify and commercialise the framework.

As 36 Kr notes, the real test will be how many developers give up the comfortable chat box and disassemble the agent into components to embed into their own products, potentially ushering in a new era of native AI applications.

Sources

  1. opensourceforu.com › OpenAI Open Sources Codex Harness Framework - Open Source For You
  2. eu.36kr.com › Shocking News: OpenAI Fully Open-Sources Codex Harness – Key Details & Impact
  3. kucoin.com › OpenAI Open-Sources Codex Harness Framework for AI Agent Development | KuCoin

More on Tech