OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn’t be the norm
OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn’t be the norm
OpenAI 应政府要求限制 GPT-5.6 发布，称此类限制不应成为常态
OpenAI is limiting the release of its newest AI models to a “small group of trusted partners” at the behest of the U.S. government, the company said Friday. The next generation GPT-5.6 lineup includes Sol, its flagship model; Terra, a more balanced model for everyday use; and Luna, a faster, lower-cost option. Although Sol is the company’s most powerful model, the Trump administration has restricted the release of all three. OpenAI said the preview is limited to partners “whose participation has been shared with the government.”
OpenAI 周五表示，应美国政府的要求，该公司正在将其最新 AI 模型的发布范围限制在“一小部分受信任的合作伙伴”内。新一代 GPT-5.6 系列包括旗舰模型 Sol、更适合日常使用的平衡模型 Terra，以及速度更快、成本更低的 Luna。尽管 Sol 是该公司最强大的模型，但特朗普政府已对这三款模型均实施了发布限制。OpenAI 表示，预览版仅限于“其参与情况已向政府报备”的合作伙伴。
The administration’s request comes as the U.S. government puts new pressure on AI companies to restrict their most advanced systems. After Anthropic released its most powerful public model Fable 5, the administration ordered the company to remove access for any foreign national, prompting Anthropic to take the model down entirely. The incident has brought up questions of how much power the government should have over AI model releases.
此次政府提出要求之际，正值美国政府向 AI 公司施加新压力，要求其限制最先进的系统。在 Anthropic 发布其最强大的公开模型 Fable 5 后，政府下令该公司禁止任何外国公民访问，导致 Anthropic 不得不将该模型彻底下架。这一事件引发了关于政府应在多大程度上掌控 AI 模型发布权的质疑。
Dean Ball, a former White House AI adviser and soon-to-be OpenAI employee, says President Trump’s recent executive order — which asks certain AI companies to voluntarily submit their most advanced models for government review up to 30 days before release — has created a de facto involuntary licensing regime for frontier AI, leading to heavy-handed restrictions. The problem compounds, Ball argues, when the government doesn’t have clearly defined safety standards, which could lead to endless launch delays that might not only give a hand to China in the AI race, but also jeopardize the billions of dollars going to AI infrastructure buildouts.
前白宫 AI 顾问、即将入职 OpenAI 的 Dean Ball 表示，特朗普总统最近签署的行政命令——要求特定 AI 公司在发布前最多 30 天自愿提交其最先进的模型供政府审查——实际上为前沿 AI 建立了一种非自愿的许可制度，导致了严苛的限制。Ball 认为，如果政府缺乏明确的安全标准，问题会进一步恶化，这可能导致无休止的发布延迟，不仅可能在 AI 竞赛中助中国一臂之力，还会危及投入到 AI 基础设施建设中的数十亿美元资金。
And while OpenAI did as the administration asked this time around, the AI firm made it clear it wasn’t happy with the arrangement. “We don’t believe this kind of government access process should become the long-term default,” reads a Friday blog post. “It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.” OpenAI called the preview a “short-term step” that will put GPT-5.6 on the path to broader availability in the coming weeks, as the company works with the administration to develop a new executive order framework on cybersecurity, as well as a “repeatable process for future model releases.”
尽管 OpenAI 此番遵从了政府的要求，但该公司明确表示对这一安排并不满意。周五的一篇博客文章写道：“我们不认为这种政府准入流程应该成为长期的默认模式。它将用户、开发者、企业、网络防御者和全球合作伙伴所需的最佳工具拒之门外。”OpenAI 将此次预览称为“短期举措”，旨在让 GPT-5.6 在未来几周内逐步实现更广泛的可用性，同时公司正与政府合作，制定关于网络安全的新行政命令框架，以及“未来模型发布的标准化流程”。
GPT-5.6 Sol specs
GPT-5.6 Sol 规格
OpenAI says GPT-5.6 Sol is its strongest model yet, with improved agentic capabilities in coding, biology, and cybersecurity. Sol introduces a “max” reasoning effort mode and an “ultra” mode that uses coordinated subagents to solve highly complex tasks (just the sort of neat trick that sends your token usage skyrocketing). GPT-5.6 excels at several benchmarks, says OpenAI, including being slightly better at coding workflows than Anthropic’s Claude Mythos 5, which the Trump administration also effectively banned this month. OpenAI says GPT-5.6 Sol is also competitive with Mythos preview but uses a third of the output tokens.
OpenAI 表示，GPT-5.6 Sol 是其迄今为止最强大的模型，在编码、生物学和网络安全方面的智能体（agentic）能力得到了提升。Sol 引入了“最大”推理努力模式和“超”模式，后者利用协同子智能体来解决高度复杂的任务（这正是那种会让你的 Token 用量飙升的巧妙技巧）。OpenAI 称，GPT-5.6 在多项基准测试中表现优异，在编码工作流方面略优于 Anthropic 的 Claude Mythos 5（该模型本月也被特朗普政府有效封禁）。OpenAI 表示，GPT-5.6 Sol 在性能上与 Mythos 预览版相当，但输出 Token 的消耗量仅为后者的三分之一。
To assuage any fears of its powerful models being unsafe, OpenAI says Sol includes its most robust security stack yet. It is, OpenAI says, heavily hardened against adversarial attacks and intentionally optimized to favor defensive cybersecurity work over offensive exploits. In other words, it’s designed to be hard to jailbreak, while prioritizing showing users how to defend against exploits, rather than how to hack into systems. OpenAI also says its safety guardrails are built directly into the core model’s behavior, rather than relying on a separate filter on top of it.
为了消除对其强大模型不安全的担忧，OpenAI 表示 Sol 包含了其迄今为止最强大的安全堆栈。OpenAI 称，该模型针对对抗性攻击进行了深度加固，并特意优化以侧重于防御性网络安全工作，而非攻击性利用。换句话说，它的设计初衷是难以被越狱，同时优先向用户展示如何防御漏洞，而不是如何入侵系统。OpenAI 还表示，其安全护栏直接构建在核心模型的行为中，而不是依赖于模型之上的独立过滤器。
The firm is likely trying to avoid the trap that caught Anthropic with Fable 5. In the brief moments when Fable 5 was available, whenever the model’s classifiers detected a high-risk topic — like cybersecurity, biology, or chemistry — it wouldn’t just block the prompt; it would route the request to an older model. The whole over-cautious flow and invisible downrouting led to many false positives and user backlash.
该公司很可能是在试图避免 Anthropic 在 Fable 5 上踩过的坑。在 Fable 5 可用的短暂时间内，每当模型的分类器检测到高风险主题（如网络安全、生物学或化学）时，它不仅会拦截提示，还会将请求路由到旧模型。这种过度谨慎的流程和隐形的降级路由导致了许多误报，并引发了用户的不满。
While the GPT-5.6 models are initially available only to a select group of partners, OpenAI plans to make them more broadly available to people using ChatGPT, Codex, and the API soon. GPT-5.6 comes in three sizes with tiered pricing: Sol costs $5 per million input tokens and $30 per million output tokens; Terra costs half that; and Luna costs $1 and $6, respectively. OpenAI says it has also improved prompt caching to make repeated prompts cheaper and more predictable.
虽然 GPT-5.6 系列模型目前仅向少数合作伙伴开放，但 OpenAI 计划很快将其更广泛地提供给 ChatGPT、Codex 和 API 用户。GPT-5.6 提供三种规格，并采用分级定价：Sol 的价格为每百万输入 Token 5 美元，每百万输出 Token 30 美元；Terra 的价格减半；Luna 的价格分别为 1 美元和 6 美元。OpenAI 表示，它还改进了提示词缓存（prompt caching），使重复提示词的成本更低且更具可预测性。