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

OpenAI Launches GPT-5.6 Family: Sol, Terra, and Luna Set New Benchmarks in Efficiency and Performance

OpenAI has officially released the GPT-5.6 family of models, marking a significant leap in frontier AI capabilities. The launch, which went live on July 9, 2026, follows a brief delay requested by the U.S. government to review the models' advanced cybersecurity features. The new family includes three tiers: the flagship Sol, the balanced Terra, and the cost-efficient Luna.

This release intensifies the competitive landscape, coming just days after Anthropic's widely praised Fable 5 launch. Early reviews highlight GPT-5.6's reliability and efficiency, though some testers note that Fable 5 may have superior raw intelligence for certain tasks.

Three Tiers for Different Needs

OpenAI has structured GPT-5.6 into three distinct capability tiers. The flagship Sol is designed for the most demanding professional work, while Terra offers a balanced option for everyday tasks, and Luna serves as the fastest, most cost-efficient model. This naming scheme—where the number denotes the generation and the name denotes the capability tier—is intended to give users clearer choices across intelligence, speed, and cost.

Pricing reflects this tiered approach. Sol is priced at $5 per million input tokens and $30 per million output tokens. Terra costs $2.50 input and $15 output, while Luna is priced at $1 input and $6 output. The models are available across ChatGPT, Codex, and the OpenAI API.

Performance Benchmarks and Competitive Landscape

GPT-5.6 Sol sets new state-of-the-art results on several key benchmarks. On the Artificial Analysis Coding Agent Index, Sol with max reasoning scored 80, outperforming Anthropic's Fable 5 by 2.8 points while using less than half the output tokens and costing about one-third less. On Agents' Last Exam, Sol scored 53.6, eclipsing Fable 5 by 13.1 points.

However, the competitive picture is nuanced. Early reviewers have noted that while GPT-5.6 excels at reliability and efficiency for everyday tasks, Anthropic's Fable 5 may possess greater raw intelligence for complex reasoning. Investor Matt Shumer noted on X that "for almost every task I tested, Fable was quite a bit better." Dan Shipper, CEO of Every, likened GPT-5.6 to a Porsche and Fable to "warp drive," suggesting each has distinct strengths.

Government Review and Delayed Release

The release of GPT-5.6 was notably delayed at the request of the Trump administration, which reviewed the models' advanced cybersecurity capabilities. OpenAI initially limited access to a small group of trusted partners, marking a shift from its previous release behavior. The company confirmed on July 8 that the models would become available on July 9, with global access expanding gradually over 24 hours.

This government review reflects a broader trend of increased scrutiny on frontier AI models. Both GPT-5.6 and Anthropic's Fable 5 underwent similar assessments, signaling a potential shift in the U.S. government's approach to AI oversight.

Performance and Early Reviews

Early reviews of GPT-5.6 have been largely positive, with many testers praising its speed, creativity, and reliability. MagicPath AI CEO Pietro Schirano called it "the best model I've ever used," while Theo Browne of T3 Chat noted its world-leading computer use capabilities. However, some reviewers, including investor Matt Shumer, expressed a preference for Anthropic's Fable 5 on certain tasks.

Dan Shipper, CEO of Every, offered a nuanced comparison: "If you need to get across the galaxy, use Fable. If you need to get around town using the best available tool for the job, use 5.6." This suggests that while Fable may excel at complex reasoning, GPT-5.6 is viewed as more reliable for everyday tasks.

Technical Architecture and New Features

GPT-5.6 introduces several architectural innovations. The flagship Sol model features a new ultra setting that coordinates multiple agents across parallel workstreams to complete complex tasks faster. This multi-agent approach, which defaults to four parallel agents, significantly shifts the score-latency frontier on demanding benchmarks like BrowseComp and Terminal-Bench 2.1.

The model also introduces Programmatic Tool Calling in the Responses API, allowing GPT-5.6 to write and run lightweight programs that coordinate tools, process intermediate results, and adapt workflows without requiring developers to script every step. This feature is Zero Data Retention (ZDR) compatible, addressing enterprise privacy concerns.

Performance Benchmarks and Competitive Analysis

GPT-5.6 Sol sets new benchmarks across multiple domains. On the Artificial Analysis Coding Agent Index, Sol with max reasoning scored 80, outperforming Anthropic's Fable 5 by 2.8 points while using less than half the output tokens and costing about one-third less. On Agents' Last Exam, Sol scored 53.6, eclipsing Fable 5 by 13.1 points.

However, the competitive landscape is nuanced. Early reviewers have noted that while GPT-5.6 excels at reliability and efficiency for everyday tasks, Anthropic's Fable 5 may possess greater raw intelligence for complex reasoning. Investor Matt Shumer noted on X that "for almost every task I tested, Fable was quite a bit better." Dan Shipper, CEO of Every, offered a memorable analogy: "If you need to get across the galaxy, use Fable. If you need to get around town using the best available tool for the job, use 5.6."

In cybersecurity, GPT-5.6 Sol achieved a 73.5% score on ExploitBench, up from GPT-5.5's 47.9%, and nearly doubled GPT-5.5's peak pass rate on ExploitGym. The model also showed strong gains in scientific research, with Pareto improvements over GPT-5.5 on life sciences evaluations like GeneBench Pro and LifeSciBench.

Government Review and Safety Measures

The release of GPT-5.6 was notably delayed at the request of the Trump administration, which reviewed the models' advanced cybersecurity capabilities. This marks a shift in the U.S. government's approach to AI oversight, as both GPT-5.6 and Anthropic's Fable 5 underwent similar assessments. OpenAI initially limited access to a small group of trusted partners before the broader release.

OpenAI has implemented its most robust safety system to date for GPT-5.6. The company conducted extensive red teaming, including approximately 700,000 A100e GPU hours of black-box automated testing. The safeguards are layered, combining protections trained into the model with real-time checks, a reasoning monitor, and account-level enforcement. For cybersecurity, a new Trusted Access program allows verified individuals and organizations to access more of the model's defensive capabilities.

Pricing and Availability

GPT-5.6 is priced per 1M tokens: Sol at $5 input / $30 output, Terra at $2.50 input / $15 output, and Luna at $1 input / $6 output. The models are available across ChatGPT, Codex, and the OpenAI API. In ChatGPT, Plus, Pro, Business, and Enterprise users can access Sol through medium and higher effort settings. The rollout began globally on July 9 and will reach full availability over 24 hours.

The launch intensifies competition among leading AI developers, coming just days after Anthropic's Fable 5 release and alongside news of an upcoming Grok 4.5 from Elon Musk's xAI. As developers begin to test GPT-5.6, the market will decide whether its efficiency and reliability advantages outweigh the raw intelligence of competing models.