Atlas

Models

← All models

GPT-5 Chat Latest (EQ-Bench 2025-10-03 label)

OpenAI

EQ-Bench labels this Spiral-Bench v1.2 run `gpt-5-chat-latest-2025-10-03`. OpenAI documents `gpt-5-chat-latest` as the non-reasoning GPT-5 model previously used in ChatGPT, but does not document a `2025-10-03` immutable snapshot, so Atlas preserves the source label as a benchmark entry.

Benchmark scores

BenchmarkScore
Spiral-Bench v1.2 Safety Score70.8
Loading Atlas data…