🔶 Altman：GPT-5.6 在 Agentic 编码任务上 token 效率提升 54%

OpenAI 首席执行官 Sam Altman 周四对 CNBC 表示，公司最新的 人工智能 模型 GPT-5.6 Sol 在 Agentic 编码任务上的 token 效率提升了 54%，并且其表现“与市场上的竞争模型不相上下，甚至更强”。

“现在每家企业都在思考支出以及他们用 AI 换取的价值，这正是我们真正想做的事情。”Altman 说。

OpenAI 周四开始推送其最新模型系列 GPT-5.6 Sol、Terra 和 Luna，该系列于上月公布。应美国政府要求，公司最初将发布范围限制在“一小群可信赖的合作伙伴”内。

Altman 表示，公司与商务部长 Howard Lutnick、财政部长 Scott Bessent 及美国国家网络总监 Sean Cairncross 就审批流程进行了合作。他将 OpenAI 与政府的协作描述为“反复沟通的合作过程”，政府会展开测试并提出需要公司解决的问题。

Altman 说：“如果你想获得广泛的访问权限，这是我们所希望的，同时你又拥有强大的模型，那你就必须对自己的安全性主张非常有信心，否则世界很快就会感到不安。”

OpenAI 正在与特朗普政府就政府可能入股事宜进行初步且持续的商讨，CNBC 此前已有报道。据 《金融时报》 报道，OpenAI 提议让政府持有 5% 的股份，但 Altman 称其中“存在许多不准确之处”。

Altman 表示，他希望监管方式能在全球范围内推行，人们可以在无需操心安全问题的前提下使用 AI。

“每个人都能用上，”他说，“并不是美国会从中获得不成比例的好处。”

OpenAI 于 2015 年作为非营利研究机构成立，并在 2022 年凭借 ChatGPT 聊天机器人一跃进入主流视野。这家如今被私募投资者估值 8520 亿美元 的公司，掀起了一场 AI 主导权争夺战，其对手包括 Anthropic、Google、Microsoft、Amazon、Meta 以及 Elon Musk 旗下的 SpaceX。

Meta 于周四发布其最新的 AI 模型 Muse Spark 1.1，并称这是其“迄今为止在 Agentic 和编码工作中最强的模型”。SpaceX 今年早些时候收购了 Musk 的 AI 初创公司 xAI，并于周三推出了名为 Grok 4.5 的模型。

SpaceX 上个月完成了 史上最大规模的 IPO，而 OpenAI 和 Anthropic 也正准备进行潜在的巨额公开市场首秀。两家公司均已 秘密提交 招股说明书，但尚未披露正式计划或时间表。

对于 OpenAI 是否会在今年 IPO，Altman 守口如瓶。

“我不知道。”他说。

视频/嵌入内容：OpenAI 首席执行官：中国的开源模型正变得非常优秀

OpenAI CEO Sam Altman told CNBC on Thursday that GPT-5.6 Sol, the company's latest artificial intelligence model, is 54% more token efficient on agentic coding tasks, and that it's "as good or better" than competing models on the market.

"Every enterprise now is thinking about spend and the value they're getting in exchange for AI, and this is what we really want to do," Altman said.

OpenAI on Thursday is rolling out its latest series of models, GPT-5.6 Sol, Terra and Luna, which it announced last month. The company limited the initial launch to a "small group of trusted partners" at the request of the U.S. government.

Altman said the company worked with Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent and U.S. National Cyber Director Sean Cairncross on the approval process. He described OpenAI's work with the government as a "collaborative back and forth," where the government would carry out tests and raise problems for the company to address.

"If you want broad access, which we do, and you have powerful models, you really want to be able to be confident in your safety claims, because otherwise the world is going to get uncomfortable very fast," Altman said.

OpenAI is in preliminary and ongoing talks with the Trump administration about a possible stake in the company, as CNBC previously reported. It has proposed giving the government a 5% holding, according to a report from the Financial Times, but Altman said there are "a lot of inaccuracies there."

Altman said he hopes the regulatory approach will be global and that people will be able to use AI without having to think about safety.

"Everybody will get access," he said. "It's not like the U.S. is going to disproportionately benefit here."

OpenAI was founded as a nonprofit research company in 2015, and it rocketed into the mainstream with the launch of its ChatGPT chatbot in 2022. The company, which is now valued at $852 billion by private investors, kick-started a sprint for AI dominance among rivals including Anthropic, Google, Microsoft, Amazon, Meta and Elon Musk's SpaceX.

Meta unveiled its latest AI model, Muse Spark 1.1, on Thursday, which it said is its "strongest model for agentic and coding work yet." SpaceX, which acquired Musk's AI startup xAI earlier this year, launched a model called Grok 4.5 on Wednesday.

SpaceX pulled off the largest IPO on record last month, and OpenAI and Anthropic are also gearing up for potentially massive public markets debuts. Both companies have confidentially filed their prospectuses with regulators, but they have not disclosed official plans or timelines.

Altman stayed mum about whether OpenAI will IPO this year.

"I don't know," he said.

WATCH: OpenAI CEO: Chinese open-source models are getting very good