OpenAI Launches GPT-5.6 (Sol, Terra, Luna): Three-Tier Model Family with Programmatic Tool Calling in Responses API

On July 9, 2026, OpenAI fully released GPT-5.6, replacing a single model with three tiers: Sol (input $5/M tok, output $30/M tok), Terra ($2.50/$15), and Luna ($1/$6). Sol scores 80 on the Artificial Analysis Coding Agent Index (2.8 points above Claude Fable 5), reaches 62.6% on OSWorld 2.0, and uses 85% fewer output tokens than Opus 4.8. For developers, the key change is Programmatic Tool Calling in the Responses API. This feature runs model-written JavaScript in an isolated V8 runtime to orchestrate tools without returning each intermediate result to the model. Clio reports a 38% reduction in prompt tokens; PlayCo reports a 63.5% reduction in total tokens. Developers using this API can significantly reduce costs and latency, particularly for complex tasks requiring multiple tool calls. However, GPT-5.6 does not lead in all metrics: Fable 5 still leads on the Artificial Analysis Intelligence Index, GDPval-AA v2, and Toolathlon, while Claude Mythos 5 leads SWE-Bench Pro by about 15 points.