GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum

Source: OpenAI Date: 2025-11-12 URL: https://openai.com/index/gpt-5-system-card-addendum-gpt-5-1

Summary

Title-only: Safety documentation addendum for GPT-5.1’s two variants — “Instant” (optimized for speed/low-latency) and “Thinking” (extended reasoning capability). The addendum format indicates this supplements the base GPT-5 system card with model-variant-specific evaluations: the thinking variant likely has distinct risk profiles from extended chain-of-thought reasoning, while the instant variant may have different capability/safety trade-offs from speed optimization.

Implications

The model variant safety thread. GPT-5.1’s variant-specific addenda reflect a new challenge: as model families proliferate (Instant, Thinking, Codex variants), safety evaluation must track the variants rather than just the base model. Different optimization targets (speed, reasoning depth, code) can shift risk profiles even within the same model family. OpenAI publishing addenda for each variant is the responsible approach but also creates significant documentation overhead.

GPT-5.x cadence. The GPT-5 → GPT-5.1 → GPT-5.2 (December 2025) update cadence suggests OpenAI is shipping continuous improvement releases rather than discrete generation jumps. This is a product model shift — more like iPhone software updates than Xbox generations. The system card addenda are the safety documentation infrastructure that makes this rapid release cadence responsible, though the pace raises questions about whether evaluations keep up with releases.