Inference
EN

GPT-5.2-Codex vs GPT-5.3 Codex

Change models
GPT-5.2-CodexOpenAI
GPT-5.3 CodexOpenAI
IndexAll categories, out of 100.
59.9 ±4.2
62.0 ±5.3 best
GapIndex points below the top.
−2.1 · no clear gap
Top
CoverageIndex weight that has results.
95% best
85%
SpeedOutput tokens per second.
44/s
52/s best
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2025-12
2026-02
Input / 1MUS dollars per 1M input tokens.
$1.75
$1.75
Output / 1MUS dollars per 1M output tokens.
$14.00
$14.00
1K chats2K tokens in, 500 out, each.
$10.50
$10.50
1K documents20K tokens in, 1K out, each.
$49.00
$49.00
1K agent tasks200K tokens in, 20K out, each.
$630.00
$630.00
ContextMaximum tokens in one request.
400K
400K
AgenticMulti-step tasks with tools.
58.1
60.6 best
CodingCode writing and repair.
61.4
61.8 best
ReasoningLogic problems and puzzles.
60.0
67.0 best
MultimodalTasks with images and text.
54.5
57.3 best
KnowledgeFacts and expert knowledge.
60.1
64.1 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
61.3 best
59.5
MathMath problems.
63.5
—

Benchmarks

12 shared
Only GPT-5.2-Codex: 10Only GPT-5.3 Codex: 9
GPT-5.2-Codex
GPT-5.3 Codex
Artificial Analysis Humanity's Last ExamKnowledge
35.7%59 / 222
42.5% best38 / 222
Artificial Analysis Long Context ReasoningReasoning
82.3%27 / 217
83.3% best13 / 217
Critical Physics TasksReasoning
8.7%73 / 214
16.9% best43 / 214
Artificial Analysis GPQA DiamondKnowledge
89.9%37 / 199
91.5% best25 / 199
τ²-Bench Tool-Agent-User EvaluationAgentic
92.1% best35 / 149
86.0%54 / 149
Artificial Analysis IFBenchInstruction
77.6% best9 / 146
75.4%23 / 146
LiveCodeBenchCoding
88.0% best9 / 142
87.3%16 / 142
Vibe Code Bench v1.1Coding
37.9%76 / 123
61.8% best58 / 123
Artificial Analysis MMMU-ProMultimodal
76.3%46 / 113
78.5% best36 / 113
SWE-benchCoding
72.4%55 / 87
78.0% best33 / 87
Gert Labs Composite Game BenchmarkAgentic
51.8%18 / 52
57.5% best13 / 52
JobBenchAgentic
26.0%28 / 36
33.7% best23 / 36