Inference
EN

Claude Haiku 5.5 vs Hy4 preview

Change models
Claude Haiku 5.5Anthropic
Hy4 previewTencent
IndexAll categories, out of 100.
66.3 ±2.8
66.3 ±4.1
GapIndex points below the top.
Top
Top
CoverageIndex weight that has results.
95% best
90%
SpeedOutput tokens per second.
159/s best
67/s
EloLMArena rating and rank.
1399
—
ReleasedMonth of release.
2026-10
2026-08
Input / 1MUS dollars per 1M input tokens.
$0.10 best
$0.834
Output / 1MUS dollars per 1M output tokens.
$0.50 best
$2.50
Batch output / 1MBatch price per 1M output tokens.
$0.25
—
1K chats2K tokens in, 500 out, each.
$0.45 best
$2.92
1K documents20K tokens in, 1K out, each.
$2.50 best
$19.18
1K agent tasks200K tokens in, 20K out, each.
$30.00 best
$216.82
ContextMaximum tokens in one request.
1M
1.05M best
AgenticMulti-step tasks with tools.
68.5 best
63.4
CodingCode writing and repair.
70.8 best
65.4
ReasoningLogic problems and puzzles.
67.9
69.4 best
MultimodalTasks with images and text.
59.5
75.5 best
KnowledgeFacts and expert knowledge.
54.4
67.7 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
37.2
—
MathMath problems.
68.2
72.9 best

Benchmarks

25 shared
Only Claude Haiku 5.5: 60Only Hy4 preview: 28
Claude Haiku 5.5
Hy4 preview
Critical Physics TasksReasoning
18.9% best33 / 214
16.9%43 / 214
Vibe Code Bench v1.1Coding
90.4% best4 / 123
77.5%37 / 123
MedScribeHealthcare
89.5% best6 / 109
83.6%45 / 109
MedCodeHealthcare
51.5% best15 / 106
43.2%47 / 106
Legal Research BenchLegal
43.8%25 / 90
45.2% best22 / 90
Tax Agent BenchFinance
31.9%27 / 82
32.6% best21 / 82
Finance Agent (v2)Finance
54.1%28 / 79
55.1% best23 / 79
Code MigrationCoding
44.6%20 / 77
47.4% best14 / 77
EMBFinance
59.0% best37 / 75
58.1%38 / 75
ProgramBenchCoding
2.5% best8 / 65
0.0%34 / 65
Terminal-Bench 4.0Agentic
36.4% best21 / 61
8.1%48 / 61
Agent Arena steerabilityAgentic
2.4 best24 / 54
0.226 / 54
Agent Arena task outcomeAgentic
1.628 / 54
3.5 best21 / 54
ProofBench v1.1Math
89.0% best12 / 54
75.0%19 / 54
Humanity's Last Exam without toolsKnowledge
45.9% best14 / 53
43.4%20 / 53
Public Benefits Bench v1.1Public benefits
63.2%25 / 50
68.6% best9 / 50
CyberBench v1.1Other
75.5% best6 / 48
65.4%32 / 48
IOICoding
47.3%29 / 45
59.3% best17 / 45
Terminal-Bench-ScienceScience
10.0% best16 / 44
1.4%37 / 44
Humanity's Last Exam with toolsAgentic
57.4% best11 / 36
55.4%18 / 36