Atlas

Models

← All models

GPT-5-Codex

OpenAI · 2025-09-15

GPT-5-Codex is OpenAI's version of GPT-5 optimized for agentic coding tasks in Codex or similar environments and trained with reinforcement learning on real-world coding tasks. OpenAI released it in Codex on September 15, 2025; the gpt-5-codex API identifier followed on September 23, 2025 and referred to a regularly updated underlying model snapshot.

Benchmark scores

BenchmarkScore
AA-LCR69.0
AA-LCR v1.171.7
AA-Omniscience Index-7.5
AIME 2025 (Artificial Analysis avg@10)98.7
Artificial Analysis Omniscience Accuracy38.7
Artificial Analysis Omniscience Hallucination Rate75.4
Artificial Analysis Openness Index5.6
Artificial Analysis Progress Index v1275
BullshitBench v2 Leaderboard39.0
BuseyBench SVG3.3
Capability149.2
CritPt (avg@5)5.1
GPQA Diamond (Artificial Analysis avg@5)83.7
Humanity's Last Exam (Text-Only)27.8
IFBench (Artificial Analysis avg@5)74.1
IOI9.8
IOI - IOI 20242.8
IOI - IOI 202516.7
LiveCodeBench (Artificial Analysis)84.0
LiveCodeBench (Vals AI public dataset)84.7
LiveCodeBench (Vals Public) - Easy98.1
LiveCodeBench (Vals Public) - Hard68.6
LiveCodeBench (Vals Public) - Medium87.5
MMLU Pro86.5
MMMU Pro73.8
Ruby LLM Benchmarks — Program Fixer64.6
Ruby LLM Program Fixer — Calendar82.8
Ruby LLM Program Fixer — Parking Garage38.9
Ruby LLM Program Fixer — School Library59.2
Ruby LLM Program Fixer — Vending Machine77.7
Loading Atlas data…