OpenAI Launches GPT-5.6: Sol, Terra, and Luna Models Now Publicly Available

Posted on 11th Jul 2026 13:04:58 in

Tagged as: OpenAI, GPT-5.6, Sol, Terra, Luna, ChatGPT, large language model, AI models, Sam Altman, artificial intelligence

Introduction

OpenAI has officially launched its GPT-5.6 family of models to the public, marking one of the most significant AI releases of 2026. The new series — comprising the flagship Sol, mid-range Terra, and budget-friendly Luna — became available to users worldwide on July 9, 2026, following a weeks-long security review with the United States government.

CEO Sam Altman announced the launch with characteristic brevity on X: "GPT-5.6 sol launches thursday! happy building." The release ends a period of restricted access that had limited the models to a select group of trusted partners since their initial preview in June, at the request of US officials concerned about the models' cybersecurity capabilities.

The launch positions OpenAI at the centre of an intensifying AI arms race, with Anthropic extending promotional access to its competing Claude Fable 5 model and Meta releasing its own Muse Spark 1.1 coding model in the same week.

The Three-Tier Model Family

OpenAI has structured the GPT-5.6 family into three distinct tiers, each targeting different use cases and budgets:

GPT-5.6 Sol is the flagship model, designed for complex reasoning, agentic workloads, and enterprise-grade applications. It is priced at $5 per million input tokens and $30 per million output tokens. Sol achieves state-of-the-art results across coding, cybersecurity, scientific research, and knowledge work. CEO Sam Altman told CNBC that Sol is "54 percent more token efficient on agentic coding tasks" compared to competing models.

GPT-5.6 Terra serves as the balanced mid-range option, delivering performance competitive with GPT-5.5 at roughly half the cost. Priced at $2.50 per million input tokens and $15 per million output tokens, Terra is positioned for everyday professional work — drafting, analysis, and business applications.

GPT-5.6 Luna is the most cost-efficient model in the lineup at $1 per million input tokens and $6 per million output tokens. Despite its lower price point, Luna outperforms Claude Opus 4.8 on key benchmarks, making it a compelling option for high-volume, budget-sensitive deployments.

All three models are available through ChatGPT, Codex, and the OpenAI API, with developers able to access them programmatically. OpenAI also offers a fast mode for Sol with speeds up to 750 tokens per second at $12.50 input and $75 output per million tokens.

Benchmark Performance and Technical Capabilities

The GPT-5.6 family sets new standards across multiple independent benchmarks. On the Artificial Analysis Coding Agent Index, Sol with maximum reasoning achieved a score of 80 — 2.8 points above Anthropic's Claude Fable 5 — while using less than half the output tokens, taking less than half the time, and costing approximately one-third less.

On Agents' Last Exam, an evaluation spanning 55 professional fields, Sol scored 53.6, eclipsing Claude Fable 5 by 13.1 points. Even at medium reasoning settings, Sol beats Fable 5 by 11.4 points at roughly one-quarter the estimated cost. Terra and Luna each outperform Fable 5 at approximately one-sixteenth the cost.

OpenAI has also introduced a new capability tier called ultra, which coordinates four AI agents in parallel by default to complete complex tasks faster. This multi-agent architecture shifts the score-latency frontier upward — reaching stronger results in less time across evaluations like BrowseComp, SEC-Bench Pro, and Terminal-Bench 2.1. Developers can build similar multi-agent experiences using the new multi-agent beta in the Responses API.

Another notable innovation is Programmatic Tool Calling, which allows the model to write and execute lightweight programs that coordinate tools, process intermediate results, and adapt workflows autonomously. Early adopters reported matching quality while using 24% fewer output tokens and completing tasks 28% faster.

Cybersecurity: OpenAI's Most Robust Model

One of the most scrutinised aspects of GPT-5.6 is its cybersecurity capability. OpenAI describes Sol as its "strongest cybersecurity model yet, achieving frontier performance with significantly fewer tokens." The model supports defensive activities including threat modelling, code review and patching, and blue teaming — simulating attacks on an organisation's own systems to identify vulnerabilities before malicious actors do.

These capabilities drew attention from the Trump administration, which initially requested that OpenAI limit the rollout to trusted partners. The US Department of Commerce ultimately approved the broad launch following technical testing and meetings between company officials and government agencies. Altman acknowledged the collaborative nature of the review, telling CNBC: "It was a collaborative back and forth, but we made many changes through the process."

The White House officially denies the process amounts to government approval, maintaining that AI companies submit models for scrutiny on a voluntary basis under the AI executive order signed by President Trump in June 2026.

GPT-Live and ChatGPT Work

Alongside the GPT-5.6 models, OpenAI launched GPT-Live, a new generation of voice models that can listen and speak simultaneously, creating a more natural conversational experience. Two versions — GPT-Live-1 and GPT-Live-1 mini — are rolling out to ChatGPT users worldwide.

The company also introduced ChatGPT Work, an enterprise-focused workspace companion that runs on desktop, web, and mobile. Designed as a daily productivity tool for teams, it can draft documents, build spreadsheets, create presentations, and generate polished interactive visualisations from natural-language prompts.

Industry Context and Competitive Response

The GPT-5.6 launch arrives in a fiercely competitive environment. Anthropic — OpenAI's primary rival — preemptively extended promotional access to Claude Fable 5, allowing users to utilise the model for up to 50% of their weekly subscription limits at no extra cost. Anthropic had faced its own regulatory hurdles: the company was forced to disable access to Claude Fable 5 and Mythos 5 models to comply with a US export control directive, which was later lifted in late June.

Meta also entered the fray this week, releasing Muse Spark 1.1 — described by CEO Mark Zuckerberg as "a strong agentic and coding model at a very low price" — and Muse Image, an AI image generator that raised immediate controversy over its ability to manipulate public Instagram users' photos.

Enterprise partners including Cursor, Qodo, Notion, Cognition, Shopify, Cisco, and Balyasny Asset Management praised GPT-5.6 in testimonials published by OpenAI, citing improvements in persistence, efficiency, and autonomous problem-solving. Simon Last, Co-Founder at Notion, described Sol as "the most tenacious problem-solver we've seen yet, staying focused and on-task for days at a time."

Design and Knowledge Work Capabilities

Beyond raw intelligence metrics, GPT-5.6 demonstrates a meaningful improvement in design judgment and computer-use capabilities. The model can inspect and refine rendered interfaces — not just generate underlying code — catching visual and functional issues before handing work back to users. On the OSWorld 2.0 benchmark for computer use, Sol scored 62.6%, surpassing Claude Opus 4.8 while using 85% fewer output tokens.

On BrowseComp, a benchmark for agentic browsing tasks, Sol achieved a new state-of-the-art score of 92.2%. These results suggest that GPT-5.6 is not merely a better text generator but a more capable autonomous operator — one that can navigate, inspect, and interact with real digital environments.

What This Means for Developers and Enterprises

The public availability of GPT-5.6 represents a significant shift for the AI development landscape. The pricing structure — with Luna at $1 per million input tokens and $6 per million output tokens — makes frontier-level AI accessible to startups and independent developers who were previously priced out of state-of-the-art models. Meanwhile, Sol's ultra mode offers large enterprises a pathway to tackle complex multi-step problems that previously required human teams coordinating over days or weeks.

OpenAI's emphasis on token efficiency also addresses a growing industry concern. As Palo Alto Networks CEO Nikesh Arora noted this week, AI pricing needs to fall dramatically as token costs continue to rise. GPT-5.6's ability to deliver stronger results with fewer tokens represents a step toward more sustainable AI economics — more intelligence per dollar, not just more intelligence at any cost.

For developers, the combination of Programmatic Tool Calling, multi-agent orchestration, and improved coding benchmarks makes GPT-5.6 a compelling upgrade. Early feedback from engineering teams at Shopify, Ramp, and Cognition suggests the model handles long-running technical workflows with markedly less steering and higher reliability than its predecessors.

Sources

• OpenAI launches its new family of models with GPT-5.6 — TechCrunch
• OpenAI to publicly release GPT-5.6, rolls out conversational AI models — CNBC
• GPT-5.6: Frontier intelligence that scales with your ambition — OpenAI (Official)
• OpenAI to release GPT-5.6 Sol, Terra and Luna on July 9 — Neowin
• OpenAI to launch GPT-5.6 after making many changes — The Straits Times