OpenAI Unveils New Flagship Models - GPT-5.6 Series

OpenAI has released its latest generation of large language models, the GPT-5.6 family, which includes three variants: Luna, Terra, and Sol (scaling from smallest to largest). These models are now generally available through the OpenAI API.

The pricing structure is per 1 million input/output tokens:

  • Luna: $1/$6
  • Terra: $2.50/$15
  • Sol: $5/$30

While price-per-token can be misleading due to varying reasoning efficiency, this provides a general comparison point.

Key Features Across All Models

All three GPT-5.6 models share these specifications:

  • Knowledge cutoff: February 16, 2026
  • Context window: 1 million tokens
  • Maximum output: 128,000 tokens

Performance Highlights

OpenAI claims GPT-5.6 demonstrates significant improvements in long-running agentic tasks. In one benchmark using Agents’ Last Exam (evaluating professional workflows across 55 fields), Sol outperformed Anthropic’s Claude Fable 5 by a substantial margin:

  • GPT-5.6 Sol: 53.6 points
  • Claude Fable 5: 40.2 points

This represents an 11.4-point advantage for Sol at roughly one-quarter the estimated cost.

Notable Strengths and Limitations

While GPT-5.6 excels in agentic reasoning, early evaluations suggest a different profile for coding tasks. For example, SWE-Bench Pro showed Claude Fable 5 achieving 80% compared to Sol’s 64.6%. This may explain OpenAI’s recent publication addressing concerns about benchmark quality in code evaluation.

New API Features

The GPT-5.6 release introduces several powerful new capabilities:

  • Programmatic Tool Calling: Enables models to orchestrate tool usage through JavaScript - potentially bridging the gap between simple prompts and complex CLI workflows.
  • Multi-Agent Systems: Allows developers to create systems with parallel, specialized subagents for more efficient task completion.
  • Prompt Cache Breakpoints: Gives users finer control over when prompts are cached (though OpenAI still supports automatic detection).
  • Improved Image Processing: The detail: original setting ensures images are processed at their native resolution without resizing.

These additions position GPT-5.6 as a significant step forward in AI capabilities, particularly for enterprise applications requiring complex workflows and integrations.