Atlas

Models

← All models

GPT-5.3-Codex

OpenAI · 2026-02-05

GPT-5.3-Codex is OpenAI's reasoning model optimized for agentic coding tasks in Codex or similar environments. Released on February 5, 2026, it combines the frontier coding performance of GPT-5.2-Codex with GPT-5.2's reasoning and professional knowledge capabilities for long-running work involving research, tool use, and complex execution.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo866
AA-Briefcase v1.1 Elo868
AA-LCR74.0
AA-LCR v1.183.3
AA-Omniscience Index10.9
ALE-Bench1934
APEX (Mercor)15.2
APEX-Agents29.9
APEX-SWE41.5
APEX-v1-extended65.9
Arena Code WebDev — Brand & Marketing1391
Arena Code WebDev — Consumer Product1402
Arena Code WebDev — Content Creation Tools1432
Arena Code WebDev — Data & Analytics1395
Arena Code WebDev — Frontend1406
Arena Code WebDev — Gaming1447
Arena Code WebDev — HTML1418
Arena Code WebDev — React1395
Arena Code WebDev — Reference-Based Design1417
Arena Code WebDev — Simulations1412
Artificial Analysis Omniscience Accuracy52.9
Artificial Analysis Omniscience Hallucination Rate89.2
Artificial Analysis Progress Index v1301
BLXBench77.7
BLXBench Coding category98.5
BLXBench Cost category92.7
BLXBench Debugging category79.6
BLXBench Hallucination category74.4
BLXBench pass rate61.2
BLXBench Reasoning category74.7
Loading Atlas data…