ClareNow

OpenAI's GPT-5.6 Is Dropping on Thursday: What's Different About Sol, Terra and Luna

OpenAI says its flagship model for ChatGPT should make fewer mistakes.

Katelyn Chedraoui CNET 08 Jul 2026, 13:44 2 min read 7/10

OpenAI's GPT-5.6 Is Dropping on Thursday: What's Different About Sol, Terra and Luna

Key Takeaways

OpenAI will release GPT-5.6 on Thursday, with three variants codenamed Sol, Terra, and Luna targeting different performance tiers.
The model is claimed to 'make fewer mistakes,' with internal benchmarks showing a 40% reduction in hallucination rates compared to GPT-4o.
A government safety review by the US AI Safety Institute must be completed before full public access, a condition not applied to previous models.
GPT-5.6 incorporates advances in chain-of-thought reasoning and reinforcement learning, with Sol reportedly scoring 93% on the MMLU benchmark.
Pricing details remain undisclosed, but enterprise beta testers indicate Sol will cost 50% more per token than GPT-4 Turbo.

OpenAI's next-generation language model, GPT-5.6, is set to launch on Thursday, promising fewer mistakes and enhanced reliability — a response to months of criticism over hallucination and safety lapses. The new flagship model for ChatGPT will debut in three distinct variants: Sol, Terra, and Luna, each tailored for different use cases. Sol is the most capable, designed for complex reasoning and enterprise apps; Terra balances performance and cost; Luna is a lightweight, fast version for real-time chat. OpenAI says the model undergoes a mandatory government review for safety before full deployment, signaling closer regulatory oversight. This release comes as competitors like Google's Gemini and Anthropic's Claude push for higher accuracy. Analysts say GPT-5.6 marks a crucial intermediate step toward GPT-5, expected in 2026, while the naming scheme suggests a shift toward planetary-scale AI infrastructure. Users can expect fewer fabricated answers and more consistent logic, but the real test will be in high-stakes fields like healthcare and law. OpenAI is also releasing a technical paper detailing improvements in reinforcement learning from human feedback and chain-of-thought reasoning. With this launch, OpenAI aims to restore trust after controversies around bias and misinformation. The company has not yet disclosed pricing tiers for Sol, Terra, and Luna, but early access partners signal premium pricing for Sol. The government review — conducted by the US AI Safety Institute — could delay the rollout by days if significant risks are found. Observers expect the new model to lift ChatGPT's subscription numbers, currently at 200 million weekly active users. The broader implication: AI safety and granular model specialization are becoming standard. GPT-5.6 is not a revolution, but it is a carefully engineered evolution designed to prove that large models can be both powerful and dependable. The next milestone to watch is the public benchmark scores, expected next week, and whether GPT-5.6 outperforms GPT-4o in all standard tests.

Frequently Asked Questions

GPT-5.6 is the latest flagship language model from OpenAI, scheduled for release on Thursday. It comes in three variants: Sol, Terra, and Luna, and is designed to produce fewer errors than previous models.

Sol, Terra, and Luna are codenames for three tiers of GPT-5.6. Sol is the most powerful variant for complex reasoning, Terra balances capability and cost, and Luna is a lightweight, fast option for real-time applications.

OpenAI plans to release GPT-5.6 on Thursday, pending completion of a government safety review by the US AI Safety Institute.

OpenAI says GPT-5.6 makes fewer mistakes through improvements in reinforcement learning from human feedback, chain-of-thought reasoning, and more accurate training data filters. Early tests show a 40% drop in hallucination compared to GPT-4o.

GPT-5.6 is the first OpenAI model subject to a mandatory safety review by the US AI Safety Institute, part of a voluntary agreement between major AI companies and the White House to ensure responsible deployment.

Yes, GPT-5.6 outperforms GPT-4 and GPT-4o on standard benchmarks like MMLU and shows significant improvements in factual accuracy and logical consistency, especially the Sol variant.

Original source

www.cnet.com

Read original

Discussion

Join the discussion

No comments yet. Be the first to share your thoughts!