Inference
EN

Hy4 vs Muse Wonder Full

Change models
Hy4Tencent
Muse Wonder FullMeta
IndexAll categories, out of 100.
72.0 ±7.8
77.3 ±5.3 best
GapIndex points below the top.
−5.3 · no clear gap
Top
CoverageIndex weight that has results.
65%
80% best
SpeedOutput tokens per second.
52/s
—
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-11
2026-10
Input / 1MUS dollars per 1M input tokens.
$0.417
—
Output / 1MUS dollars per 1M output tokens.
$1.25
—
1K chats2K tokens in, 500 out, each.
$1.46
—
1K documents20K tokens in, 1K out, each.
$9.59
—
1K agent tasks200K tokens in, 20K out, each.
$108.41
—
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
74.5
78.2 best
CodingCode writing and repair.
68.4
77.3 best
ReasoningLogic problems and puzzles.
—
75.6
MultimodalTasks with images and text.
57.4
—
KnowledgeFacts and expert knowledge.
83.7 best
71.7
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
—
MathMath problems.
—
79.6

Benchmarks

11 shared
Only Hy4: 10Only Muse Wonder Full: 31
Hy4
Muse Wonder Full
Vibe Code Bench v1.1Coding
81.0%30 / 123
86.0% best14 / 123
Legal Research BenchLegal
41.3%30 / 90
53.4% best6 / 90
Tax Agent BenchFinance
34.4%17 / 82
43.0% best8 / 82
Terminal-Bench 4.0Agentic
39.4%15 / 61
58.1% best6 / 61
DeepSWEAgentic
74.9%8 / 59
79.0% best1 / 59
Agent Arena steerabilityAgentic
8.814 / 54
9.6 best12 / 54
Agent Arena task outcomeAgentic
6.712 / 54
11.1 best9 / 54
HLAB-Vals-1.1Legal
12.5%17 / 35
23.3% best5 / 35
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding
30.0%12 / 19
59.0% best9 / 19
OSWorld 2.1 (all 108 tasks)Agentic
78.9% best5 / 11
78.4%6 / 11
SWE-MarathonCoding
43.8%
57.5%