Inference
EN

DeepSeek V4 Flash 0731 vs GPT-5.3 Codex

Change models
DeepSeek V4 Flash 0731DeepSeek
GPT-5.3 CodexOpenAI
IndexAll categories, out of 100.
61.9 ±3.8
62.0 ±5.3 best
GapIndex points below the top.
−0.1 · no clear gap
Top
CoverageIndex weight that has results.
85%
85%
SpeedOutput tokens per second.
61/s best
52/s
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-07
2026-02
Input / 1MUS dollars per 1M input tokens.
$0.043 best
$1.75
Output / 1MUS dollars per 1M output tokens.
$0.30 best
$14.00
Batch output / 1MBatch price per 1M output tokens.
$0.144
—
1K chats2K tokens in, 500 out, each.
$0.237 best
$10.50
1K documents20K tokens in, 1K out, each.
$1.17 best
$49.00
1K agent tasks200K tokens in, 20K out, each.
$14.68 best
$630.00
ContextMaximum tokens in one request.
1.05M best
400K
AgenticMulti-step tasks with tools.
59.9
60.6 best
CodingCode writing and repair.
62.8 best
61.8
ReasoningLogic problems and puzzles.
65.5
67.0 best
MultimodalTasks with images and text.
—
57.3
KnowledgeFacts and expert knowledge.
58.6
64.1 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
59.2
59.5 best
MathMath problems.
61.5
—

Benchmarks

11 shared
Only DeepSeek V4 Flash 0731: 74Only GPT-5.3 Codex: 10
DeepSeek V4 Flash 0731
GPT-5.3 Codex
Artificial Analysis Humanity's Last ExamKnowledge
38.6%53 / 222
42.5% best38 / 222
Artificial Analysis Long Context ReasoningReasoning
79.7%45 / 217
83.3% best13 / 217
Critical Physics TasksReasoning
16.6%46 / 214
16.9% best43 / 214
Artificial Analysis GPQA DiamondKnowledge
90.8%30 / 199
91.5% best25 / 199
LiveCodeBenchCoding
87.3%16 / 142
87.3%16 / 142
Vibe Code Bench v1.1Coding
74.7% best43 / 123
61.8%58 / 123
Software Engineering Benchmark VerifiedCoding
79.0%24 / 114
85.0% best9 / 114
SWE-benchCoding
88.8% best11 / 87
78.0%33 / 87
SWE-bench ProCoding
52.6%63 / 83
56.8% best46 / 83
IOICoding
32.7%40 / 45
53.8% best22 / 45
Terminal-Bench 2.0 · BenchLMAgentic
56.9%23 / 41
77.3% best2 / 41