OpenAI’s Codex team took to Reddit on July 10 for an AMA session about GPT-5.6, the company’s newest model family that started rolling out just one day earlier. The flagship model, dubbed Sol, arrived alongside two siblings, Terra and Luna, each targeting different price points and performance levels.
The Sol model sits at the top of the lineup, scoring an 80 on the Artificial Analysis Coding Agent Index, which puts it at or near state-of-the-art territory for coding benchmarks.
Pricing follows a tiered structure. Luna comes in at $1 per million input tokens and $6 per million output tokens. Terra sits in the middle at $2.50 and $15 respectively. Sol commands $5 input and $30 output per million tokens. A 90% read discount on prompt caching makes repeated queries significantly cheaper for production workloads.
The Codex integration now lives directly inside the ChatGPT desktop application, with support spanning repositories, terminals, browsers, and mobile platforms. OpenAI reported that weekly Codex users have grown to over 5 million, doubling in just three months.
Among the additions is a multi-agent “ultra” mode designed for complex task execution, with multiple agents coordinating to tackle different parts of a problem simultaneously. OpenAI also introduced programmatic tool calling, which lets the model interact with external systems more reliably, alongside advanced safeguards meant to keep autonomous coding agents from going rogue.
