OpenAI's three-tier GPT-5.6 lineup launched Thursday is winning praise from Chinese users who bypass the country's ChatGPT block via VPNs, even as the models cost more than local alternatives from Zhipu AI and DeepSeek. The family includes Sol, the flagship model tuned for coding and cybersecurity research; Terra, delivering GPT-5.5-class performance at roughly half the price of that predecessor; and Luna, a lightweight option built for low inference cost. OpenAI also introduced an "ultra" setting that runs multiple agents in parallel for higher performance.
Sol is priced at $5 per million input tokens and $30 per million output tokens. Terra costs $2.50 and $15, respectively. Luna, the entry-level option, runs $1 for input tokens and $6 for output tokens.
Those prices undercut OpenAI's own previous models but still exceed Chinese rivals by a wide margin. Zhipu AI's GLM-5.2 charges about $1.40 per million input tokens and $4.40 per million output tokens. DeepSeek V4 costs up to $0.44 per million input tokens and $0.87 for output.
Despite the premium, Chinese users accessing the blocked service via VPNs and third-party proxies have praised GPT-5.6's efficiency, according to the South China Morning Post. The release shows a industry shift toward cost-efficient AI systems designed for enterprise use, Li Yitao, a Chinese entrepreneur and co-founder of Canada-based AI startup Quotaflow, told the outlet.
Access to the three models opened Thursday in ChatGPT and Codex, with Sol available to Plus, Pro, Business, and Enterprise subscribers. Free and Go users get GPT-5.6 Terra instead.
OpenAI claims GPT-5.6 delivers "a step change in design judgment" over past models and excels at creative applications including interactive data visualizations. The company also says the new model blocks ten times as much nefarious activity compared to earlier safeguards, a response to concerns about AI-powered vulnerability discovery following Anthropic's Fable 5 headlines last month. For internal development, OpenAI says GPT-5.6 is having its biggest impact yet when used by its own engineers to build future models.
