GPT-6.1 Terra 系统卡增补：生物化学能力定为 High，Sandbox Bench 通过率 22.7%

2. Model Data and Training

GPT-6.1 Terra uses the same types of data and training as GPT-6 Astra, described in the GPT-6 Astra card.

For previously launched models, the values published at launch reflect the versions evaluated at that time. The comparison values for previously launched models that are shown here may reflect later versions of those models, and may vary from the values published at launch.¹

---

1. GPT-6.1 Terra is intended to be used in accordance with OpenAI’s Usage Policies, Service Terms, and Terms of Use. These policies apply universally to OpenAI services and are designed to ensure safe and responsible usage of AI technology. You can review OpenAI’s Usage Policies at openai.com/policies/usage-policies/ .

   If you need assistance with respect to these models, you can find further information on OpenAI’s website (openai.com), or you can contact OpenAI Support by opening the chat bubble icon displayed at the bottom-right of help.openai.com. The input and output modalities supported by the models can be found at https://developers.openai.com/api/docs/models/all.

   A list of the languages that ChatGPT currently supports can be found here.↩︎

9.1 Biological and Chemical Capabilities

We are treating GPT-6.1 Terra as High capability in the Biological and Chemical domain. This is illustrated by Multimodal Troubleshooting Virology crossing its indicative threshold. The four available Critical capability results do not cross their listed thresholds.

For a full description of these evaluations, please see the Biological and Chemical Capabilities section in the GPT-6 Astra card.

9.2.3 Sandbox Bench

Sandbox Bench is an internal evaluation designed to test a model’s ability to identify and exploit vulnerabilities in runtimes, file format parsers, and egress proxies inspired by those used in internal systems. While the security of these software libraries and systems is not load bearing to our security threat model – for example, runtimes are further contained by containers, operating systems and hardened virtual machines – this evaluation is meant to provide signal on areas where investing in first-layer hardening will improve security in a defense in depth approach.

GPT-6.1 Terra succeeded on 5 of 22 targets (22.7%), compared to 9 of 22 targets (41.9%) for GPT-6.1 Sol and compared to 10 of 22 targets (45.5%) by GPT-6 Astra.

9.4.1 Model Safety Training and Evaluation

9.4.1.1 Biological and Chemical Safety Training and Evaluation

GPT‑6.1 Terra performs on par with recent GPT-6 Astra and GPT-6.1 releases. These results reflect model responses alone, without our full production safeguards.

Table 31. Biology Model Refusal Evaluation (higher is better)

9.4.1.2 Cybersecurity Safety Training and Evaluation

GPT‑6.1 Terra performs on par with recent GPT-6 Astra and GPT-6.1 releases. These results reflect model responses alone, without our full production safeguards.

Table 32. Cybersecurity Safety Evaluation (higher is better)