Inference
EN

DeepSeek V4 Pro 0813 vs Qwen3.8 Max

Change models
DeepSeek V4 Pro 0813DeepSeek
Qwen3.8 MaxAlibaba
IndexAll categories, out of 100.
65.5 ±3.7
65.6 ±2.5 best
GapIndex points below the top.
−0.1 · no clear gap
Top
CoverageIndex weight that has results.
85%
95% best
SpeedOutput tokens per second.
42/s best
41/s
EloLMArena rating and rank.
—
1485
ReleasedMonth of release.
2026-08
2026-08
Input / 1MUS dollars per 1M input tokens.
$0.111
—
Output / 1MUS dollars per 1M output tokens.
$4.44
—
Batch output / 1MBatch price per 1M output tokens.
$1.98
—
1K chats2K tokens in, 500 out, each.
$2.44
—
1K documents20K tokens in, 1K out, each.
$6.66
—
1K agent tasks200K tokens in, 20K out, each.
$111.00
—
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
63.7
63.7
CodingCode writing and repair.
65.1 best
64.3
ReasoningLogic problems and puzzles.
68.5 best
66.7
MultimodalTasks with images and text.
—
67.2
KnowledgeFacts and expert knowledge.
65.0 best
62.6
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
62.0
71.4 best
MathMath problems.
64.1
67.6 best

Benchmarks

51 shared
Only DeepSeek V4 Pro 0813: 36Only Qwen3.8 Max: 62
DeepSeek V4 Pro 0813
Qwen3.8 Max
GPQA diamond · Epoch AIKnowledge
91.7%30 / 217
92.7% best25 / 217
OTIS Mock AIME 2024-2025Math
98.6%27 / 200
99.4% best17 / 200
LegalBenchLegal
82.4%72 / 149
83.6% best51 / 149
TaxEval v2Finance
73.1%52 / 143
75.6% best14 / 143
LiveCodeBenchCoding
87.5%13 / 142
87.9% best10 / 142
Graduate-Level Google-Proof Q&AKnowledge
90.1%29 / 141
92.6% best15 / 141
GPQA Diamond · Vals AIKnowledge
92.4%16 / 135
93.7% best6 / 135
MMLU ProKnowledge
87.0%35 / 135
88.6% best18 / 135
CorpFin v2Finance
65.4%27 / 131
65.9% best25 / 131
Vibe Code Bench v1.1Coding
82.3% best25 / 123
64.7%55 / 123
MedScribeHealthcare
80.2%59 / 109
84.9% best33 / 109
MedCodeHealthcare
42.5% best50 / 106
40.7%61 / 106
SimpleQA VerifiedKnowledge
52.9% best26 / 91
45.8%41 / 91
Chess PuzzlesReasoning
46.0%26 / 90
46.0%26 / 90
FrontierMath-Tiers-1-3-v2-PrivateMath
64.6%39 / 90
74.7% best24 / 90
Legal Research BenchLegal
40.9%33 / 90
47.6% best15 / 90
SWE-benchCoding
96.4% best2 / 87
85.6%17 / 87
Terminal-Bench 2.1 · BenchLMAgentic
87.9% best12 / 86
86.6%15 / 86
GPQA Diamond · BenchLMKnowledge
90.1%30 / 84
92.6% best17 / 84
SWE-bench ProCoding
55.4%54 / 83
67.7% best11 / 83