# State of AI 2026 (Clone)

## Main Cover
GPT-5.6
Sol
Preview
OpenAI · June 26, 2026

## Overview
Overview · Product Release
OpenAI is beginning a limited preview of the **GPT-5.6 series**: **Sol**, its flagship model; **Terra**, a balanced model for everyday work; and **Luna**, a fast and affordable model.
Terra has competitive performance to GPT-5.5 while being 2x cheaper, and Luna brings strong capability at the company's lowest cost.
GPT-5.6 Sol launches with OpenAI's most robust safety stack to date. The company strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse, spending multiple weeks finding weaknesses, pressure-testing the system, and hardening it against real-world attacks.
As part of its ongoing engagement with the U.S. government, OpenAI previewed its plans and the models' capabilities ahead of launch. At the government's request, the rollout is starting with a limited preview for a small group of trusted partners before releasing more broadly.
OpenAI plans to make Sol, Terra, and Luna generally available in the coming weeks as it continues testing and coordinating with partners.
GPT-5.6 Series · Preview

## Contents
Table of Contents
Inside the preview
| No. | Section | Description | Color | Face |
| --- | --- | --- | --- | --- |
| 01 | The GPT-5.6 series | Sol, Terra and Luna | #ff7a00 | family |
| 02 | Capabilities | New max and ultra reasoning modes | #ff9f1c | landscape |
| 03 | Terminal-Bench 2.1 | A new state of the art in coding | #ffb000 | grouped-bars |
| 04 | Safety & access | Stronger cyber capability, stronger safeguards | #ffc83d | talent |
| 05 | Availability | Limited preview, GA in coming weeks | #e0a020 | availability |
GPT-5.6 Series · Preview

## The Series
THE GPT-5.6 SERIES
Three models, one release
| Name | Tier | Description | Color |
| --- | --- | --- | --- |
| Sol | Flagship | OpenAI's strongest model yet, with new max and ultra reasoning for the deepest, most complex work. | #ff7a00 |
| Terra | Balanced | Competitive performance with GPT-5.5 for everyday work, while being 2x cheaper. | #ff9f1c |
| Luna | Fast & affordable | Strong capability delivered at OpenAI's lowest cost, built for speed and scale. | #ffc83d |
GPT-5.6 Series

## Capabilities
Capabilities
01
GPT-5.6 Sol is OpenAI's strongest model yet. It introduces a new max reasoning effort that gives Sol the most time to reason deeply, plus a new ultra mode that goes beyond a single agent by leveraging subagents to accelerate complex work. Improvements span coding, biology, and cybersecurity.
GPT-5.6 Sol

## Terminal-Bench
BENCHMARK · CODING
A new state of the art on Terminal-Bench 2.1
Tests command-line workflows requiring planning, iteration, and tool coordination.
| Model | Score (%) | Highlight |
| --- | --- | --- |
| GPT-5.6 Sol Ultra | 91.9 | Yes |
| GPT-5.6 Sol | 88.8 | Yes |
| GPT-5.5 | 88.0 | No |
| GPT-5.6 Luna | 84.3 | Yes |
| Claude Mythos 5 | 84.3 | No |
| Claude Fable 5 | 83.4 | No |
| GPT-5.6 Terra | 82.5 | Yes |
| Claude Opus 4.8 | 78.9 | No |
| Gemini 3.1 Pro Preview | 70.7 | No |
Terminal-Bench 2.1, OpenAI, June 26, 2026

## Capability Highlights
Stronger across coding, biology and cyber
| Number | Label | Description |
| --- | --- | --- |
| 91.9% | Terminal-Bench 2.1 | new state of the art for coding, set by GPT-5.6 Sol Ultra |
| ~1/3 | output tokens | on ExploitBench, while staying competitive with Mythos Preview |
| 2 | new reasoning modes | max effort for deeper thinking and ultra for subagent work |
On Terminal-Bench 2.1, Sol sets a new state of the art for command-line workflows that require planning, iteration, and tool coordination.
On GeneBench v1, Sol shows broad gains in long-horizon genomics and quantitative-biology analyses, beating GPT-5.5 while using fewer tokens.
On ExploitBench, Sol is competitive with Mythos Preview using only about a third of the output tokens, shifting the performance-efficiency frontier for security work.
On ExploitGym, a benchmark from UC Berkeley with OpenAI and other frontier labs, Sol, Terra and Luna all improve as reasoning increases.

## Pull Quote
SAFETY PRINCIPLE
Sol is better at helping people <em>find and fix</em> vulnerabilities than reliably carrying out end-to-end attacks. OpenAI's stated priority is to make sure these capabilities reach and benefit defenders.
OpenAI
From the GPT-5.6 Sol preview announcement
GPT-5.6 SOL

## Safety & Access
Safety & access
02
Sol is OpenAI's most capable model yet for cybersecurity, so it ships with the company's most robust safeguards to date. The goal is to make prohibited offensive activity more difficult, uncertain, and detectable, while preserving legitimate work like code review, patch development, and defensive testing. Sol does not cross the Cyber Critical threshold under OpenAI's Preparedness Framework.
GPT-5.6 Sol

## Safeguard Stack
A LAYERED SAFEGUARD STACK
How OpenAI hardened the release
| Index | Heading | Description |
| --- | --- | --- |
| 01 | Capability-matched safeguards | Protections are configured to each model's capabilities across Sol, Terra and Luna. |
| 02 | Protect, don't block | Legitimate work like vulnerability research, patching, debugging and security education stays open. |
| 03 | Adversarial pressure-testing | Weeks of finding weaknesses, pressure-testing the system, and hardening against real-world attacks. |
| 04 | Automated red-teaming | Continuous robustness testing constrains prohibited offensive use while benefiting defenders. |
GPT-5.6 SOL

## Availability
AVAILABILITY & PRICING
A limited preview, then broad access
| Heading | Description |
| --- | --- |
| Limited preview now | Starting with a small group of trusted partners, shared with the U.S. government, before a broader release. |
| Generally available soon | OpenAI plans to make Sol, Terra and Luna broadly available in the coming weeks. |
| Terra: 2x cheaper | Competitive performance with GPT-5.5 for everyday work at half the cost. |
| Luna: lowest cost | Strong capability delivered at our most affordable price point. |
OpenAI says it believes in broad access, framing this short-term step as the strongest path to making these models widely available.
GPT-5.6 Series · OpenAI