GPT-5.3 (HalluHard label)
OpenAI
HalluHard reports a generative `gpt-5.3` entry, and its public evaluation registry maps that label to an OpenAI Responses API request using model `gpt-5.3` at medium reasoning effort, with or without web search. OpenAI publicly documents `gpt-5.3-chat-latest` (GPT-5.3 Instant) and `gpt-5.3-codex`, but not a released base `gpt-5.3` artifact, so Atlas retains this source-specific identity as a benchmark entry.
Benchmark scores
| Benchmark | Score |
|---|---|
| HalluHard (All Domains Average) | 43.5 |