The new GPT-5.6 family: Luna, Terra, Sol

2026-07-09 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, short

Summary

OpenAI released its new GPT-5.6 model family, Luna, Terra, and Sol, on July 9th, 2026, available in three sizes. Pricing per 1M input/output tokens is \$1/\$6 for Luna, \$2.50/\$15 for Terra, and \$5/\$30 for Sol. OpenAI claims significant improvements in long-running agentic performance, with GPT-5.6 Sol achieving 53.6 on the Agents' Last Exam, surpassing Claude Fable 5 by 13.1 points. The smaller Luna and Terra models also outperform Fable 5 at significantly lower costs. However, GPT-5.6 Sol scored 64.6% on SWE-Bench Pro, while Fable 5 achieved 80%, prompting OpenAI to question SWE-Bench Pro's reliability. The release also introduces new API features: Programmatic Tool Calling for orchestrating JavaScript tool calls, Multi-agent support for parallel subagents, Prompt cache breakpoints for explicit caching, and a "detail: original" option for image requests to prevent resizing.

Key takeaway

For AI Engineers evaluating large language models for agentic workflows, GPT-5.6 offers compelling performance and cost efficiency on long-running tasks. You should investigate its new API features like Programmatic Tool Calling and Multi-agent capabilities to streamline complex automation. Consider testing its performance against your specific coding benchmarks, as general evaluations like SWE-Bench Pro may not fully reflect real-world utility.

Key insights

OpenAI's GPT-5.6 models prioritize agentic efficiency and introduce advanced API features for complex workflows.

Principles

• Agentic performance can significantly reduce operational costs.
• Benchmark reliability requires careful scrutiny and auditing.
• Explicit cache control offers optimization opportunities.

Method

The models can compose and run JavaScript for tool orchestration and spin up subagents for parallel task execution via new API features.

In practice

• Explore Programmatic Tool Calling for CLI utility orchestration.
• Implement Multi-agent patterns directly through the API.
• Utilize Prompt cache breakpoints for cost-optimized API calls.

Topics

• GPT-5.6
• Agentic AI
• LLM API Features
• Programmatic Tool Calling
• Multi-agent Systems
• LLM Benchmarking

Best for: CTO, VP of Engineering/Data, Investor, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• OpenAI's GPT-5.5 is here, and it's no potato: narrowly beats Anthropic's Claude Mythos Preview on Terminal-Bench 2.0 — GPT-5.5 redefines AI agency, excelling in complex, multi-part tasks with less human guidance.
• OpenAI launches its new family of models with GPT-5.6 — OpenAI's GPT-5.6 models offer significant efficiency and performance gains, particularly in coding and cybersecurity, challenging competitors.
• So I've been using gpt-5.6 for awhile... — GPT 5.6 demonstrates unprecedented capability for complex, long-running, multi-step tasks, including extensive code rewrites and autonomous computer interaction.
• 😺 OpenAI shipped GPT-5.5 today — GPT-5.5's release intensifies AI model competition, showcasing a shift towards task-oriented capabilities and a potential "soul swap" in model characteristics.
• I Tried The New GPT 5.5 And I’m Never Going Back — GPT 5.5 excels at complex task execution and agentic workflows, moving beyond simple query responses.
• GPT-5.6 Pro LEAKED & Is Coming Soon! Mythos 5 Level! — GPT 5.6 Pro elevates LLM performance through enhanced reasoning, integrated tooling, and sophisticated code and simulation generation.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.