Inference
EN

Hy4 preview vs Inkling 2 Small

Change models
Hy4 previewTencent
Inkling 2 SmallThinking Machines Lab
IndexAll categories, out of 100.
66.4 ±4.1
66.4 ±3.8
GapIndex points below the top.
Top
Top
CoverageIndex weight that has results.
90%
90%
SpeedOutput tokens per second.
67/s
—
EloLMArena rating and rank.
—
1403
ReleasedMonth of release.
2026-08
2026-11
Input / 1MUS dollars per 1M input tokens.
$0.834
$0.30 best
Output / 1MUS dollars per 1M output tokens.
$2.50
$1.00 best
1K chats2K tokens in, 500 out, each.
$2.92
$1.10 best
1K documents20K tokens in, 1K out, each.
$19.18
$7.00 best
1K agent tasks200K tokens in, 20K out, each.
$216.82
$80.00 best
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
63.5
67.6 best
CodingCode writing and repair.
65.5
66.6 best
ReasoningLogic problems and puzzles.
69.6 best
68.4
MultimodalTasks with images and text.
75.7 best
63.6
KnowledgeFacts and expert knowledge.
67.9 best
61.5
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
—
MathMath problems.
73.0 best
60.4

Benchmarks

22 shared
Only Hy4 preview: 31Only Inkling 2 Small: 36
Hy4 preview
Inkling 2 Small
Graduate-Level Google-Proof Q&AKnowledge
92.3%18 / 141
93.0% best11 / 141
Vibe Code Bench v1.1Coding
77.5%37 / 123
81.5% best28 / 123
Legal Research BenchLegal
45.2%22 / 90
46.6% best18 / 90
Terminal-Bench 2.1 · BenchLMAgentic
85.4% best18 / 86
85.1%19 / 86
GPQA Diamond · BenchLMKnowledge
92.3%19 / 84
93.0% best12 / 84
Tax Agent BenchFinance
32.6%21 / 82
38.9% best11 / 82
Humanity's Last ExamKnowledge
55.4% best14 / 77
53.2%16 / 77
ProgramBenchCoding
0.0%34 / 65
0.5% best19 / 65
Terminal-Bench 4.0Agentic
8.1%48 / 61
12.6% best41 / 61
DeepSWEAgentic
64.3%36 / 59
72.1% best17 / 59
Agent Arena steerabilityAgentic
0.226 / 54
9.3 best13 / 54
Agent Arena task outcomeAgentic
3.5 best21 / 54
3.323 / 54
Humanity's Last Exam without toolsKnowledge
43.4%20 / 53
44.1% best17 / 53
MCP AtlasAgentic
83.7%8 / 52
85.0% best5 / 52
CyberBench v1.1Other
65.4%32 / 48
72.7% best16 / 48
Terminal-Bench-ScienceScience
1.4%37 / 44
2.9% best33 / 44
JobBenchAgentic
61.7%9 / 36
70.5% best2 / 36
SRE BenchOther
2.3% best24 / 36
1.9%25 / 36
HLAB-Vals-1.1Legal
8.8%23 / 35
21.7% best6 / 35
BioMysteryBenchScience
69.3%15 / 30
70.0% best14 / 30