Hy3 Preview vs Kimi K2.5

Change models
Hy3 PreviewTencent
Kimi K2.5Moonshot AI
IndexAll categories, out of 100.
52.3 ±7.4 best
52.2 ±2.2
GapIndex points below the top.
Top
−0.1 · no clear gap
CoverageIndex weight that has results.
75%
100% best
SpeedOutput tokens per second.
69/s best
36/s
EloLMArena rating and rank.
1401
—
ReleasedMonth of release.
2026-04
2026-01
Input / 1MUS dollars per 1M input tokens.
$0.18 best
$0.532
Output / 1MUS dollars per 1M output tokens.
$0.60 best
$3.32
1K chats2K tokens in, 500 out, each.
$0.66 best
$2.73
1K documents20K tokens in, 1K out, each.
$4.20 best
$13.97
1K agent tasks200K tokens in, 20K out, each.
$48.00 best
$172.90
ContextMaximum tokens in one request.
262K
262K
AgenticMulti-step tasks with tools.
52.6 best
49.4
CodingCode writing and repair.
55.2 best
53.9
ReasoningLogic problems and puzzles.
49.7
51.6 best
MultimodalTasks with images and text.
—
52.4
KnowledgeFacts and expert knowledge.
53.9 best
53.7
MultilingualTasks in many languages.
—
52.7
InstructionTasks with strict prompt rules.
48.0
53.5 best
MathMath problems.
—
53.8

Benchmarks

13 shared
Only Hy3 Preview: 2Only Kimi K2.5: 76
Hy3 Preview
Kimi K2.5
33.5% best68 / 222
30.7%74 / 222
66.7%132 / 217
78.0% best70 / 217
4.9% best84 / 214
3.1%95 / 214
89.7% best39 / 199
87.9%53 / 199
31.5%75 / 164
35.2% best65 / 164
87.2%43 / 141
87.6% best41 / 141
35.8% best49 / 133
17.0%83 / 133
74.4%48 / 114
76.8% best38 / 114
87.2%48 / 84
87.6% best47 / 84
25.5%58 / 77
30.1% best48 / 77
36.9%43 / 52
45.9% best29 / 52
54.4% best25 / 41
50.8%28 / 41
41.2%18 / 35
48.7% best9 / 35