OpenAI Officially Launches GPT-5.6 Family and Unveils ChatGPT Work
모델 출시 | Fri Jul 10 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 6 sources
OpenAI released three models—Sol, Terra, and Luna—alongside Microsoft 365 Copilot integration and the ChatGPT Work agent.
Analysis
[OpenAI] officially released the GPT-5.6 model family [2][5]
- Comprised of three tiers: flagship Sol
- general-purpose Terra
- and low-cost Luna
- Supports parallel multi-agent workstreams via ultra setting
- Achieved SOTA in coding
- cybersecurity
- and science domains
- Enhanced computer use and design judgment
[GPT-5.6 Sol] set SOTA records on coding and agent benchmarks [2][4][5]
- Scored 80 on the Artificial Analysis Coding Agent Index
- Outperformed Claude Fable 5 by 2.8 points while using less than half the output tokens
- Scored 53.6 on Agents' Last Exam
- 13.1 points above Fable 5
- 54% token efficiency improvement on AI coding tasks
[OpenAI API] published pricing for the three GPT-5.6 models [4][6]
- Sol: $5 input / $30 output per million tokens
- Terra: $2.50 input / $15 output
- Luna: $1 input / $6 output
- Available across ChatGPT
- Codex
- and the OpenAI API
[ChatGPT Work] unveiled a workplace AI agent for non-developers [3][4]
- A work agent combining ChatGPT and Codex capabilities
- Integrated plugin directory connecting Slack
- Gmail
- Google Drive
- calendar
- and CRM
- Automatically generates documents
- spreadsheets
- presentations
- and web apps
- Immediately available via Mac and Windows desktop apps
- including for free users
[Microsoft 365 Copilot] adopted GPT-5.6 as its default preferred model [1]
- Integrated across Word
- Excel
- PowerPoint
- Chat
- and Cowork
- Direct model access via the OpenAI API
- Improved quality in document writing
- data analysis
- and presentations
- Supports automation of cross-functional collaboration
[OpenAI Safety] strengthened cybersecurity and safeguards for GPT-5.6 [2][3][4]
- Public rollout proceeded after Trump administration approval
- Supports threat modeling
- code review
- and blue team defensive activities
- Combined human red teaming with large-scale automated testing
- Real-time monitoring and trust-based access controls
Sources
- [1] GPT-5.6 is now the preferred model in Microsoft 365 Copilot - OpenAI Blog
- [2] GPT-5.6: Frontier intelligence that scales with your ambition - OpenAI Blog
- [3] OpenAI rolls out GPT-5.6 after government greenlight — and announces ‘ChatGPT Work’ - The Verge AI
- [4] OpenAI launches its new family of models with GPT-5.6 - TechCrunch AI
- [5] GPT-5.6 - Hacker News
- [6] Anthropic Wants You to Pay Up for Claude Fable 5 - Wired AI