OpenAI releases GPT-5.6 with Sol, Terra, and Luna models, integrates Codex into ChatGPT

OpenAI released GPT-5.6 on July 9, 2026, introducing three distinct model tiers under a new naming system: Sol, Terra, and Luna. The company also integrated its Codex development tool directly into the main ChatGPT application.

Three-tier model system

The GPT-5.6 release marks a shift in OpenAI's model naming convention:

• Sol: OpenAI's flagship model for the 5.6 generation
• Terra: Balanced model designed for everyday work
• Luna: Fast and affordable model

According to OpenAI, "the number identifies a model's generation, while Sol, Terra, and Luna identify durable capability tiers that can advance on their own cadence." This system allows each tier to receive updates independently rather than requiring synchronized releases across all models.

Codex integration

OpenAI merged its Codex development environment into the main ChatGPT application, a feature the company had previously discussed publicly. The desktop ChatGPT app now includes Codex-powered capabilities.

Specific technical details about the Codex integration, including supported programming languages or IDE features, were not disclosed in the announcement.

ChatGPT Work and hosting features

The company launched ChatGPT Work, available across web, mobile, and desktop platforms. OpenAI also introduced a hosted sites feature for all paid users, though pricing and technical specifications for these services were not announced.

Missing technical specifications

OpenAI has not yet disclosed:

• Context window sizes for Sol, Terra, or Luna
• Pricing per 1M tokens for any of the three models
• Benchmark scores (MMLU, HumanEval, etc.)
• Parameter counts
• Training data cutoff dates
• Specific performance differences between the three tiers

The announcement was made during a livestream at 10 AM PT featuring Andrew Ambrosino, Jessica Liang, Ed Bayes, Lauren Gordon, Tejal Patwardhan, and Katy Shi.

What this means

OpenAI's three-tier system represents a structural change in how the company positions its models, emphasizing choice across "intelligence, speed, and cost" rather than a single flagship offering. The independent update cadence for each tier could allow faster iteration on specific use cases without requiring full model generation releases. However, without pricing or benchmark data, it remains unclear how these models compare to competitors like Anthropic's Claude 3.5 Sonnet or whether the performance-cost tradeoffs justify the added complexity of choosing between three options.