Inference
EN

MiniCPM5-1B vs MiniCPM5-2B

Change models
MiniCPM5-1BOpenBMB
MiniCPM5-2BOpenBMB
IndexAll categories, out of 100.
15.5 ±7.0
37.1 ±5.0 best
GapIndex points below the top.
−21.6
Top
CoverageIndex weight that has results.
75%
85% best
SpeedOutput tokens per second.
—
—
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-05
2026-09
Input / 1MUS dollars per 1M input tokens.
—
—
Output / 1MUS dollars per 1M output tokens.
—
—
1K chats2K tokens in, 500 out, each.
—
—
1K documents20K tokens in, 1K out, each.
—
—
1K agent tasks200K tokens in, 20K out, each.
—
—
ContextMaximum tokens in one request.
131K
131K
AgenticMulti-step tasks with tools.
8.5
38.2 best
CodingCode writing and repair.
11.9
32.0 best
ReasoningLogic problems and puzzles.
—
39.1
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
9.2
34.7 best
MultilingualTasks in many languages.
5.0
—
InstructionTasks with strict prompt rules.
28.9
44.4 best
MathMath problems.
21.9
43.9 best

Benchmarks

12 shared
Only MiniCPM5-1B: 6Only MiniCPM5-2B: 12
MiniCPM5-1B
MiniCPM5-2B
GPQA Diamond · BenchLMKnowledge
26.3%83 / 84
70.2% best72 / 84
Massive Multitask Language Understanding ProfessionalKnowledge
48.9%66 / 75
70.8% best55 / 75
Instruction-Following EvalInstruction
80.4%54 / 70
86.7% best35 / 70
LiveCodeBench v6Coding
33.5%57 / 59
69.1% best42 / 59
Instruction Following BenchmarkInstruction
46.7%52 / 57
66.3% best42 / 57
American Invitational Mathematics Examination 2025Math
40.4%46 / 53
86.5% best25 / 53
AIME 2026Math
40.4%33 / 34
86.5% best28 / 34
Harvard-MIT Mathematics Tournament February 2026Math
25.8%27 / 27
63.8% best25 / 27
MATH-500 Problem SetMath
91.6%19 / 27
94.6% best17 / 27
Berkeley Function Calling Leaderboard v4Agentic
25.1%24 / 26
66.6% best11 / 26
SuperGPQA: Scaling LLM Evaluation Across 285 Graduate DisciplinesKnowledge
23.1%22 / 22
40.8% best21 / 22
MMLU-ReduxKnowledge
70.1%19 / 19
84.7% best17 / 19