OpenAI Introduces GPT-5.6 with Sol, Terra, and Luna Models, Featuring Programmatic Tool Calling

OpenAI has officially launched its GPT-5.6 model family, moving it to general availability after a limited preview. This release introduces not one, but three distinct tiers: Sol, Terra, and Luna, each designed to cater to different performance and cost requirements. A key advancement in this release is the integration of programmatic tool calling within the Responses API, allowing models to orchestrate tools more efficiently by executing JavaScript in isolated runtimes.

This development is particularly relevant for developers and businesses in India relying on advanced AI tools for automation, coding, and complex workflows. The new capabilities aim to reduce token usage and improve the overall efficiency of AI-driven applications.

Model Tiers and Pricing

The GPT-5.6 family is structured into three tiers, each with specific pricing per 1 million tokens. Sol, the flagship model, is priced at $5 for input and $30 for output. Terra, designed as a balanced everyday option, costs $2.50 for input and $15 for output. Luna, the most cost-efficient tier, is available at $1 for input and $6 for output. These tiered offerings provide flexibility for various application scales and budget considerations, which could significantly impact the adoption rates among Indian startups and enterprises.

Key facts:

Model Tier	Input Price (per 1M tokens)	Output Price (per 1M tokens)
Sol	$5	$30
Terra	$2.50	$15
Luna	$1	$6

Performance Benchmarks

GPT-5.6 Sol demonstrates notable performance gains across several benchmarks. It achieves a score of 80 on the Artificial Analysis Coding Agent Index v1.1, surpassing Claude Fable 5 by 2.8 points. OpenAI reports that Sol accomplishes this with less than half the output tokens and processing time compared to its competitor. Furthermore, Sol sets new state-of-the-art results on Terminal-Bench 2.1 and DeepSWE, reaching 92.2% on BrowseComp and 62.6% on OSWorld 2.0. On OSWorld, it outperforms Claude Opus 4.8 while using 85% fewer output tokens.

For long-running professional workflows, Sol scores 53.6 on Agents’ Last Exam, an evaluation across 55 fields. This score is reported to be 13.1 points higher than Claude Fable 5’s adaptive reasoning capabilities. Such improvements in coding and complex task execution are critical for Indian AI developers working on sophisticated applications.

Programmatic Tool Calling

The substantive developer-centric change in GPT-5.6 is the introduction of programmatic tool calling. This feature allows the model to write and execute JavaScript code in an isolated V8 runtime to orchestrate various tools. This eliminates the need to return every intermediate result to the model, leading to a significant reduction in prompt tokens. Early reports indicate a 38% cut in prompt tokens for Clio and 63.5% fewer total tokens for PlayCo, illustrating the efficiency gains for developers integrating multiple tools.

This capability streamlines the development of complex AI agents and automated systems, which is highly beneficial for Indian tech companies aiming to build more efficient and autonomous solutions in areas like customer service, data analysis, and operational automation.

Market Positioning and Competition

While GPT-5.6 shows strong performance in specific areas, the competitive landscape remains dynamic. Claude Fable 5 still leads in the Artificial Analysis Intelligence Index, GDPval-AA v2, and Toolathlon. Additionally, Claude Mythos 5 holds a lead of approximately 15 points on SWE-Bench Pro. These competitive dynamics mean that developers will need to carefully evaluate which model best suits their specific needs, weighing performance against cost and specialized capabilities.

The release also includes changes to prompt caching, with GPT-5.6 supporting explicit cache breakpoints and a 30-minute minimum cache life. Cache writes are billed at 1.25x the model’s uncached input rate, while cache reads benefit from a 90% cached-input discount. These caching mechanisms are designed to optimize costs for repeated prompts.

Source: MarkTechPost, https://www.marktechpost.com/2026/07/09/openai-releases-gpt-5-6-a-three-tier-model-family-with-programmatic-tool-calling/