Inference
EN

Claude Opus 5 vs Hy4

Change models
Claude Opus 5Anthropic
Hy4Tencent
IndexAll categories, out of 100.
73.8 ±1.9 best
72.0 ±7.8
GapIndex points below the top.
Top
−1.8 · no clear gap
CoverageIndex weight that has results.
100% best
65%
SpeedOutput tokens per second.
48/s
52/s best
EloLMArena rating and rank.
1502
—
ReleasedMonth of release.
2026-07
2026-11
Input / 1MUS dollars per 1M input tokens.
$5.00
$0.417 best
Output / 1MUS dollars per 1M output tokens.
$25.00
$1.25 best
Batch output / 1MBatch price per 1M output tokens.
$12.50
—
1K chats2K tokens in, 500 out, each.
$22.50
$1.46 best
1K documents20K tokens in, 1K out, each.
$125.00
$9.59 best
1K agent tasks200K tokens in, 20K out, each.
$1,500.00
$108.41 best
ContextMaximum tokens in one request.
1M
1.05M best
AgenticMulti-step tasks with tools.
71.6
74.5 best
CodingCode writing and repair.
74.8 best
68.4
ReasoningLogic problems and puzzles.
75.4
—
MultimodalTasks with images and text.
72.5 best
57.4
KnowledgeFacts and expert knowledge.
73.1
83.7 best
MultilingualTasks in many languages.
63.6
—
InstructionTasks with strict prompt rules.
55.6
—
MathMath problems.
74.8
—

Benchmarks

13 shared
Only Claude Opus 5: 137Only Hy4: 8
Claude Opus 5
Hy4
Vibe Code Bench v1.1Coding
88.4% best11 / 123
81.0%30 / 123
Legal Research BenchLegal
55.3% best2 / 90
41.3%30 / 90
Tax Agent BenchFinance
47.1% best5 / 82
34.4%17 / 82
Humanity's Last ExamKnowledge
64.7% best4 / 77
59.1%7 / 77
Terminal-Bench 4.0Agentic
53.5% best10 / 61
39.4%15 / 61
DeepSWEAgentic
68.8%24 / 59
74.9% best8 / 59
Agent Arena steerabilityAgentic
4.417 / 54
8.8 best14 / 54
Agent Arena task outcomeAgentic
10.2 best10 / 54
6.712 / 54
Humanity's Last Exam with toolsAgentic
64.7% best3 / 36
59.1%10 / 36
AutomationBenchAgentic
26.0%31 / 32
56.2% best4 / 32
FrontierCode 2.0Coding
26.7%13 / 27
31.0% best9 / 27
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding
93.0% best1 / 19
30.0%12 / 19
OSWorld 2.1 (all 108 tasks)Agentic
74.0%8 / 11
78.9% best5 / 11