OpenAI Unveils Two Major Updates: GPT‑5.6 Launch and ChatGPT‑Codex Integration Cutting Costs Below Fable‑5

Overview

OpenAI released GPT‑5.6 today, introducing three model variants—Sol, Terra and Luna—across ChatGPT, Codex and the API. The models are already being rolled out gradually to users worldwide.

Performance and Cost Advantages

In the “Agents' Last Exam” benchmark, GPT‑5.6 Sol scored 53.6, 13.1 points higher than Claude Fable 5’s adaptive version, and outperformed Fable 5 by 11.4 points on medium‑intensity reasoning while costing roughly one‑quarter of Fable 5. Terra and Luna also beat Fable 5, with costs about one‑sixteenth of its.

On the Artificial Analysis Coding Agent Index, Sol achieved a record 80.0, 2.8 points above Fable 5, using less than half the output tokens, less than half the latency, and about one‑third the cost.

Ultra Mode and Multi‑Agent Parallelism

GPT‑5.6 ships with an “Ultra” mode that coordinates multiple agents to work in parallel on difficult tasks, trading additional tokens for faster, higher‑quality results. This reflects OpenAI’s shift from single‑model answers to collaborative multi‑agent execution for long‑chain tasks.

Enhanced Design Judgment

Sol improves computer‑operation capabilities, allowing it to inspect and adjust rendered outputs, catch visual or functional issues before delivery, and generate higher‑quality PPT, document and spreadsheet artifacts that fit existing templates and can be exported to professional tools.

Rollout and Access Tiers

GPT‑5.6 is being pushed to ChatGPT, Codex and the OpenAI API, with full coverage expected within 24 hours. In ChatGPT, Plus, Pro, Business and Enterprise plans can select Sol at medium‑or‑higher inference strength; Pro and Enterprise also offer GPT‑5.6 Pro for the most demanding tasks.

The tiered rollout mirrors OpenAI’s usual strategy: different entry points, plans and inference strengths map to different capability ceilings, making the specific tier and conditions critical for developers.

ChatGPT Work: Integrated Agent Product

ChatGPT Work combines Codex’s execution engine with ChatGPT, enabling cross‑application and cross‑file actions that can run for hours, decompose goals into steps, and produce deliverables such as tables, slides, documents and web apps.

While Codex originally targeted developers, over 5 million weekly users now employ it for non‑coding tasks. The new product inherits GPT‑5.6’s high‑quality multi‑step reasoning and template‑driven generation.

Typical use cases include analyzing monthly budget variances, turning research briefs into marketing decks, and iteratively refining outputs while maintaining context.

Implications for Enterprise Users

The updates move AI from answering questions toward acting as an autonomous work system that can read/write files, coordinate agents, and deliver finished results, accelerating AI penetration into enterprise workflows.

FAQ Highlights

Core change in GPT‑5.6: simultaneous improvements in performance, speed, cost and multi‑agent execution, with Sol handling high‑difficulty, long‑chain tasks.

Difference between ChatGPT Work and previous ChatGPT: focus shifts from Q&A to sustained task completion across apps and files.

Reason for merging Codex into ChatGPT: to bring programming‑agent capabilities to a mainstream work entry point.

Model roles: Sol for hardest, longest‑chain tasks; Terra for everyday production; Luna for cheap, high‑frequency lightweight jobs.

Enterprise impact: AI can now take over entire workflows such as spreadsheets, documents, PPTs, and marketing briefs.