GPT-5.6 Launches Globally, Codex Merges, and ChatGPT Work Boosts Productivity

After a limited preview, OpenAI officially released the GPT-5.6 model family on Thursday. The lineup consists of the flagship Sol model for general work, the cost‑effective Terra model, and the high‑speed Luna model. All three are available in ChatGPT, Codex and the OpenAI API and will be rolled out globally within 24 hours.

The pricing scheme is tiered: Sol costs $5 per M input tokens and $30 per M output tokens; Terra costs $2.50 / $15; Luna costs $1 / $6. A new prompt‑cache mechanism offers explicit cache breakpoints, a minimum retention of 30 minutes, and charges cache writes at 1.25 × the uncached input rate while reads retain a 90 % discount.

Performance improvements are highlighted with concrete benchmark results. On the Agents’ Last Exam, Sol achieved a new high of 53.6, surpassing Claude Fable 5 by 13.1 points while using roughly a quarter of the estimated cost. Terra and Luna also outperformed Fable 5, with costs about one‑sixteenth of the baseline. In coding evaluations, Sol set a new best of 80 on the Artificial Analysis Coding Agent Index v1.1, beating Fable 5 by 2.8 points while generating less than half the output tokens and running in less than half the time; Terra and Luna showed similar advantages over competing models.

Additional tests such as Terminal‑Bench 2.1, DeepSWE, BrowseComp, OSWorld 2.0 and the Artificial Analysis Intelligence Index confirm Sol’s superior token efficiency, reduced latency, and higher scores across 55 domains, often within 1 point of the top competitor and with 61 % faster task completion.

OpenAI also introduced a multi‑agent parallelism feature: the default ultra configuration runs four agents concurrently, improving the score‑vs‑latency curve on BrowseComp, SEC‑Bench Pro and Terminal‑Bench 2.1. Developers can access this via the beta multi‑agent Responses API.

Alongside the model release, OpenAI launched ChatGPT Work , an intelligent‑agent tool that merges ChatGPT with Codex. It can directly operate on applications and files, run for extended periods, and transform end‑to‑end goals into completed work. The service is available to Pro, Enterprise and Edu plans now, with Plus and Business plans slated to follow.

ChatGPT Work supports a wide range of use cases: converting raw data into marketing briefs, generating slides or documents, automating Slack or Teams updates, and even performing scheduled tasks that monitor dashboards or email feedback. The desktop app now includes a built‑in browser for seamless web‑based workflows, and plugins connect the agent to services such as Google Drive, SharePoint, CRM systems, and more.

OpenAI reports that over 5 million weekly users are leveraging Codex, and more than 1 million have applied it beyond code assistance, indicating a shift toward a “super‑app” era of productivity. The company notes that token consumption for ChatGPT Work differs from standard chat requests, reflecting the higher complexity and longer runtimes of the tasks it handles.

In summary, the GPT‑5.6 release combines higher model capabilities, cost‑effective pricing, advanced caching, and multi‑agent execution, while ChatGPT Work extends these models into a full‑stack productivity platform that integrates with existing tools, plugins, and scheduled automation.