BenchAtlas

Rankings / OpenAI

GPT 5 Codex

released 2025-09-23

BenchAtlas Index

as of 2026-07-14
73.5
high effort
rank #46
9 families · 5 categories · high

Benchmark evidence

11 results
agentic coding
τ²-Benchhigh effort86.8 %independentArtificial Analysis
coding
LiveCodeBench v6
implementation=artificial-analysis
high effort84.0 %independentArtificial Analysis
SciCodehigh effort40.9 %independentArtificial Analysis
external indices
Intelligence Index v4.1high effort36.1 pointsindependentArtificial Analysis
AA Math Indexhigh effort98.7 pointsindependentArtificial Analysis
knowledge science
GPQA Diamond
implementation=artificial-analysis
high effort83.7 %independentArtificial Analysis
Humanity's Last Exam
implementation=artificial-analysis
high effort25.6 %independentArtificial Analysis
MMLU-Pro
implementation=artificial-analysis
high effort86.5 %independentArtificial Analysis
long context instruction
AA-LCRhigh effort69.0 %independentArtificial Analysis
IFBenchhigh effort74.2 %independentArtificial Analysis
reasoning math
AIME
year=2025 · implementation=artificial-analysis
high effort98.7 %independentArtificial Analysis

Agent + model results

systems, not bare-model scores
agent + model Codex CLI + GPT 5 CodexTerminal-Bench 2.044.3 %communityTerminal-Bench Leaderboard
agent + model terminus-2 + GPT 5 CodexTerminal-Bench 2.043.4 %communityTerminal-Bench Leaderboard
agent + model Codex CLI + GPT 5 CodexTerminal-Bench 1.042.8 %communityTerminal-Bench Leaderboard
agent + model mini-SWE-agent + GPT 5 CodexTerminal-Bench 2.041.4 %communityTerminal-Bench Leaderboard
agent + model Artificial Analysis harness + GPT 5 CodexTerminal-Bench Hard37.9 %independentArtificial Analysis

These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.