[DigitalToday reporter Jinju Hong] OpenAI has made its new flagship artificial intelligence (AI) model, 'GPT-5.6 Sol', available to general users, moving fully into the next-generation AI race. As it revamps its brand from number-based model names to the 'Sol·Terra·Luna' system, it is targeting the market with both performance and price competitiveness.

On July 9 (local time), blockchain media outlet Decrypt reported that OpenAI ended the limited preview service for GPT-5.6 Sol that it had run for about 2 weeks and released the full version. It also unveiled the mid-tier model 'Terra' and the entry-level model 'Luna'.

The core of the announcement is a revamp of the model lineup. OpenAI applied the names Sol, Terra and Luna by performance tier instead of distinguishing models by numbers such as GPT-5.5 and GPT-5.6. Sol is the top-performing flagship model. Terra is an everyday model that offers GPT-5.5-level performance at about half the cost. Luna is positioned as an entry-level model focused on cost efficiency.

Pricing is also differentiated by performance tier. Sol costs $5 per 1 million input tokens and $30 per 1 million output tokens. Luna is priced at $1 and $6, respectively. Sol also adds a 'Max Thinking' option that increases reasoning time to improve complex problem-solving and an 'Ultra' mode that distributes tasks across multiple sub-agents.

Benchmark performance was also disclosed. In Terminal-Bench 2.1, which evaluates command-line-based task execution, Sol Ultra scored 91.9 percent and standard Sol scored 88.8 percent. That exceeds Anthropic's Claude Mythos 5 at 88.0 percent, Claude Fable 5 at 84.3 percent, Claude Opus 4.8 at 78.9 percent and Google's Gemini 3.1 Pro Preview at 70.7 percent.

OpenAI also highlighted performance in cybersecurity. In ExploitBench, which evaluates software vulnerability detection and exploitation capabilities, Sol showed a similar level of performance to Anthropic's restricted Mythos preview, but used about one-third as many tokens, it explained. However, it said its internal safety evaluation did not reach the 'cyber critical' level.

Early users' assessments are broadly positive. Theo, a developer and chief executive of T3 Chat, said it was "world-class" in computer-use tasks and that various issues he felt with GPT-5.5 had been resolved.

Dan Shipper of AI collaboration platform Every shared results from 1 month of testing and said, "GPT-5.6 is like a Porsche, and Claude Fable 5 is like a warp drive." He said Sol showed balanced performance in everyday knowledge work and coding.

Japanese AI researcher Daichi Konno also said it was clearly improved from GPT-5.5 and was difficult to rank against Claude Fable 5. He added that Anthropic still had an edge in writing and pointed out that safeguards did not work sufficiently on life-science-related questions.

The release comes as competition in the AI industry has reached its peak. Sol appeared the day after xAI unveiled 'Grok 4.5' and shortly after Meta announced 'Muse Spark 1.1'. Over the past week, OpenAI, Anthropic, xAI and Meta have all released new AI models, intensifying the next-generation generative AI race. Google's Gemini 3, however, has not yet been replaced with a new flagship model since its unveiling in November 2025.

Its price competitiveness is also notable. Anthropic's Claude Fable 5 charges $10 per 1 million input tokens and $50 for output, while xAI's Grok 4.5 charges $15 and $75, respectively. By contrast, Chinese model DeepSeek V4 Pro is priced at $1.74 and $3.48, and Xiaomi Mimo v2.5 Pro is around $1 and $5, pursuing low-price strategies. Sol is positioned strategically between high performance and reasonable pricing.

OpenAI has built a new flagship AI lineup that balances performance and cost through this model revamp. With rivals also rolling out next-generation models in quick succession, a key market watch point will be whether Sol can maintain an edge in corporate adoption and competition to secure users.

Apparently I'm allowed to talk about GPT-5.6 now? It's a damn good model. Not quite as "smart" as Fable, but it is incredibly capable. Fixed all the problems I had with GPT-5.5. It is incredibly determined. Will run for a day without even using a /goal. It understands subagents…