Inference
EN

Claude Sonnet 5.5 vs Hy4

Change models
Claude Sonnet 5.5Anthropic
Hy4Tencent
IndexAll categories, out of 100.
77.2 ±2.6 best
72.0 ±7.8
GapIndex points below the top.
Top
−5.2 · no clear gap
CoverageIndex weight that has results.
95% best
65%
SpeedOutput tokens per second.
107/s best
52/s
EloLMArena rating and rank.
1469
—
ReleasedMonth of release.
2026-09
2026-11
Input / 1MUS dollars per 1M input tokens.
$2.00
$0.417 best
Output / 1MUS dollars per 1M output tokens.
$10.00
$1.25 best
Batch output / 1MBatch price per 1M output tokens.
$5.00
—
1K chats2K tokens in, 500 out, each.
$9.00
$1.46 best
1K documents20K tokens in, 1K out, each.
$50.00
$9.59 best
1K agent tasks200K tokens in, 20K out, each.
$600.00
$108.41 best
ContextMaximum tokens in one request.
1M
1.05M best
AgenticMulti-step tasks with tools.
77.5 best
74.5
CodingCode writing and repair.
80.8 best
68.4
ReasoningLogic problems and puzzles.
77.8
—
MultimodalTasks with images and text.
74.4 best
57.4
KnowledgeFacts and expert knowledge.
71.3
83.7 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
41.5
—
MathMath problems.
76.1
—

Benchmarks

13 shared
Only Claude Sonnet 5.5: 115Only Hy4: 8
Claude Sonnet 5.5
Hy4
Vibe Code Bench v1.1Coding
92.4% best2 / 123
81.0%30 / 123
Legal Research BenchLegal
48.1% best12 / 90
41.3%30 / 90
Tax Agent BenchFinance
42.0% best10 / 82
34.4%17 / 82
Terminal-Bench 4.0Agentic
64.1% best3 / 61
39.4%15 / 61
DeepSWEAgentic
71.0%21 / 59
74.9% best8 / 59
Agent Arena steerabilityAgentic
14.5 best7 / 54
8.814 / 54
Agent Arena task outcomeAgentic
19.5 best3 / 54
6.712 / 54
Humanity's Last Exam with toolsAgentic
64.5% best4 / 36
59.1%10 / 36
HLAB-Vals-1.1Legal
10.4%20 / 35
12.5% best17 / 35
FrontierCode 2.0Coding
41.0% best4 / 27
31.0%9 / 27
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding
79.7% best6 / 19
30.0%12 / 19
OSWorld 2.1 (all 108 tasks)Agentic
80.1% best3 / 11
78.9%5 / 11
Terminal-Bench-Science 0.1Agentic
59.9% best4 / 11
19.8%11 / 11