Inference
EN

Claude Opus 4.6 (Adaptive) vs GPT-5.2-Codex

Change models
Claude Opus 4.6 (Adaptive)Anthropic
GPT-5.2-CodexOpenAI
IndexAll categories, out of 100.
60.0 ±4.7 best
59.9 ±4.2
GapIndex points below the top.
Top
−0.1 · no clear gap
CoverageIndex weight that has results.
95%
95%
SpeedOutput tokens per second.
—
44/s
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-02
2025-12
Input / 1MUS dollars per 1M input tokens.
—
$1.75
Output / 1MUS dollars per 1M output tokens.
—
$14.00
1K chats2K tokens in, 500 out, each.
—
$10.50
1K documents20K tokens in, 1K out, each.
—
$49.00
1K agent tasks200K tokens in, 20K out, each.
—
$630.00
ContextMaximum tokens in one request.
1M best
400K
AgenticMulti-step tasks with tools.
63.5 best
58.1
CodingCode writing and repair.
60.0
61.4 best
ReasoningLogic problems and puzzles.
61.0 best
60.0
MultimodalTasks with images and text.
58.8 best
54.5
KnowledgeFacts and expert knowledge.
62.6 best
60.1
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
38.6
61.3 best
MathMath problems.
57.3
63.5 best

Benchmarks

11 shared
Only Claude Opus 4.6 (Adaptive): 16Only GPT-5.2-Codex: 11
Claude Opus 4.6 (Adaptive)
GPT-5.2-Codex
Artificial Analysis Humanity's Last ExamKnowledge
39.9% best47 / 222
35.7%59 / 222
Artificial Analysis Long Context ReasoningReasoning
78.0%70 / 217
82.3% best27 / 217
Critical Physics TasksReasoning
12.6% best56 / 214
8.7%73 / 214
Artificial Analysis GPQA DiamondKnowledge
89.6%41 / 199
89.9% best37 / 199
Artificial Analysis Omniscience AccuracyKnowledge
47.0% best30 / 164
41.1%45 / 164
τ²-Bench Tool-Agent-User EvaluationAgentic
92.1%35 / 149
92.1%35 / 149
Artificial Analysis IFBenchInstruction
53.1%76 / 146
77.6% best9 / 146
LiveCodeBenchCoding
84.7%36 / 142
88.0% best9 / 142
Vibe Code Bench v1.1Coding
53.5% best63 / 123
37.9%76 / 123
Artificial Analysis MMMU-ProMultimodal
75.4%49 / 113
76.3% best46 / 113
SWE-benchCoding
78.2% best30 / 87
72.4%55 / 87