OpenAI released GPT-5.6, a new family of models designed to raise the bar on AI reasoning and speed. The lineup includes Luna, Terra, and Sol, priced from $1 to $5 per million input tokens. All three models support a 1 million token context window and feature a knowledge cutoff from February 2026.

On benchmarks for long-running agentic tasks, the models reportedly outperform Claude Fable 5. OpenAI emphasized improved API capabilities including programmatic tool calling, multi-agent orchestration, and prompt cache breakpoints.

What GPT-5.6 Brings

The three model sizes allow developers and businesses to choose based on budget and latency requirements. A bigger context window means more information fed into a single request—useful for analyzing large documents, codebases, or conversation histories without splitting the input.

The pricing structure undercuts rival offerings. That’s intentional. OpenAI is competing aggressively for enterprise AI spending as multiple vendors release capable models.

OpenAI’s Broader Push

The same month, OpenAI upgraded ChatGPT voice mode to GPT-Live, enabling simultaneous listening and speaking. The old walkie-talkie style forced turn-taking. GPT-Live handles interruptions and real-time translation naturally.

OpenAI also rolled out ChatGPT Work, a desktop agent for project delegation. Instead of asking a question and waiting for text back, users can delegate tasks and let the agent run them end-to-end on their desktop.

These releases signal OpenAI’s shift from chat interfaces to agentic systems that can plan, execute, and iterate autonomously.