OpenAI 的 GPT-5.6 系列模型将于周四发布。 这些模型于 6 月底亮相，最初在美国政府的压力下仅限部分合作伙伴使用。据 Axios 报道，在人工智能标准与创新中心进行了额外测试后，美国商务部批准了公开发布。OpenAI 公开批评了这一限制，称其让开发者和企业无法获得最好的工具。正如特朗普最新的 AI 行政令所呼吁的那样，针对此类模型发布的具有约束力的标准至今仍不存在。
OpenAI 表示，Sol 在多项基准测试中击败了 Anthropic 的 Claude Mythos 5。在 TerminalBench 2.1 上，Sol 得分 88.8%，Sol Ultra 达到 91.9%，Mythos 5 则为 88%。在网络安全任务上，Sol 与 Mythos 5 持平，但仅使用了三分之一的 token。Sol 的价格为每百万输入/输出 token 5 美元/30 美元。Anthropic 的 Fable 5 价格几乎翻倍，为 10 美元/50 美元，而且 很可能消耗的 token 也更多。
OpenAI's GPT-5.6 models ship Thursday. Unveiled in late June, they were initially restricted to select partners under U.S. government pressure. The Department of Commerce approved the public launch after the Center for AI Standards and Innovation ran additional tests, Axios reported. OpenAI openly criticized the hold, saying it kept the best tools away from developers and companies. Binding standards for releasing such models, as called for in Trump's latest AI executive order, still don't exist.
OpenAI says Sol beats Anthropic's Claude Mythos 5 on several benchmarks. On TerminalBench 2.1, Sol scored 88.8 percent, Sol Ultra hit 91.9 percent, and Mythos 5 landed at 88 percent. On cybersecurity tasks, Sol matched Mythos 5 but used only a third of the tokens. Sol costs $5/$30 per million input/output tokens. Anthropic's Fable 5 runs nearly double at $10/$50 and likely burns through more tokens too.