Inference360
Moonshot AI

Kimi K2.5

Kimi K2.5 is a non-reasoning model from Moonshot AI. 44 benchmarks count toward its score, in 7 categories.

availableShows if the model has enough results for an index.
IndexOverall score out of 100.53.3 ±2.9
CoverageShare of the index weight with results.95%
SpeedOutput tokens per second.47/s
Input / 1MUS dollars per 1M input tokens.$0.45
Output / 1MUS dollars per 1M output tokens.$2.25
ContextMaximum tokens in one request.262K
EloLMArena rating and rank.1446 (#53)

The index is a score out of 100. The ± range shows how much it can change.

70,513 votes. Elo shows what people prefer. It does not change the score.

CapabilitiesScore per category, out of 100.

Out of 100
AgenticMulti-step tasks with tools.
49.8
CodingCode writing and repair.
54.9
ReasoningLogic problems and puzzles.
50.0
MultimodalTasks with images and text.
57.2
KnowledgeFacts and expert knowledge.
56.1
MultilingualTasks in many languages.
N/A
InstructionTasks with strict rules in the prompt.
57.0
MathMath problems.
54.8

Results

44 counted
BenchmarkThe test name.CategoryThe capability that the test measures.ResultThe score from the publisher.IndexThis result as a score out of 100.RunThe settings of the run.DateDate of the result.Published byThe source of the result.
AIME25 first-party comparison snapshotMath96.3%———Arcee AI
American Invitational Mathematics Examination 2025Math96.1%52.2——Mathematical Association of America
τ²-Bench Tool-Agent-User EvaluationAgentic95.9%67.6——Victor Barres et al.
AIME 2026Math95.8%54.8——Qwen
Harvard-MIT Mathematics Tournament February 2025Math95.4%53.8——Qwen
Instruction-Following EvalInstruction93.9%53.6——Jeffrey Zhou et al.
OTIS Mock AIME 2024-2025Math92.2%63.1——Epoch AI
Harvard-MIT Mathematics Tournament November 2025Math91.1%———Qwen
Artificial Analysis GPQA DiamondKnowledge87.9%58.6——Artificial Analysis
Graduate-Level Google-Proof Q&AKnowledge87.6%59.1——David Rein et al.
GPQA DiamondKnowledge87.6%59.1——David Rein et al.
GPQA diamondKnowledge87.6%59.1——Epoch AI
Video-MMEMultimodal87.4%———Video-MME benchmark team
Massive Multitask Language Understanding ProfessionalKnowledge87.1%57.8——Yubo Wang et al.
MMLU-Pro first-party comparison snapshotKnowledge87.1%57.8——Arcee AI
Harvard-MIT Mathematics Tournament February 2026Math87.1%55.3——Qwen
VideoMMMUMultimodal86.6%———Qwen
LiveCodeBench v6Coding85.0%54.4——LiveCodeBench maintainers
MMLU-ProXMultilingual82.3%———MMLU-ProX authors
MMAnswerBenchMath81.8%———Qwen
Multimodal Multi-disciplinary Video UnderstandingMultimodal80.4%———MMVU benchmark maintainers
Massive Multi-discipline Multimodal Understanding ProMultimodal78.5%55.1——MMMU-Pro authors
Artificial Analysis Long Context ReasoningReasoning78.0%62.2——Artificial Analysis
React Native EvalsCoding77.2%55.6——Callstack
DeepSearchQAAgentic77.1%59.3——Meta AI
Software Engineering Benchmark VerifiedCoding76.8%59.1——Carlos E. Jimenez et al.
Artificial Analysis MMMU-ProMultimodal75.4%59.2——Artificial Analysis
SWE-Bench verifiedCoding73.8%56.7——Epoch AI
WideResearchAgentic72.7%57.9——Qwen
SWE-bench Verified (mini-swe-agent-v2)Coding70.8%54.3——Arcee AI
SWE-bench VerifiedCoding70.8%54.3high effort · mini-SWE-agent1 Sept 2026SWE-bench team
Artificial Analysis IFBenchInstruction70.2%60.3——Artificial Analysis
SuperGPQA: Scaling LLM Evaluation Across 285 Graduate DisciplinesKnowledge69.2%54.8——Xiaoxuan Du et al.
SWE-bench MultilingualMultilingual67.3%—mini-SWE-agent20 Feb 2026SWE-bench team
SWE-bench MultilingualCoding67.3%—mini-SWE-agent2 Sept 2026SWE-bench team
τ³-Bench Tool-Agent-User EvaluationAgentic65.7%51.0——Sierra Research
ARC-AGI-1 (semi-private)Reasoning65.3%58.1——ARC Prize Foundation
LongBench v2Reasoning61.0%———LongBench v2 authors
BrowseCompAgentic60.6%50.8——OpenAI
MCP-TasksAgentic59.1%———Qwen
SWE-RebenchCoding58.5%———Nebius
NOVA-63Multilingual56.0%———Qwen
QwenClawBenchAgentic54.3%53.6——Qwen
Claw-EvalAgentic52.3%40.7——Bowen Ye et al.
SWE-bench ProCoding50.7%53.0——Xiang Deng et al.
Scientific Code BenchmarkCoding48.7%59.2——Benchmark authors
Artificial Analysis Coding IndexCoding46.8%51.9——Artificial Analysis
Gert Labs Composite Game BenchmarkAgentic45.9%53.3——Gert Labs
Artificial Analysis Omniscience AccuracyKnowledge35.2%57.4——Artificial Analysis
SimpleQA VerifiedKnowledge34.3%53.3——Epoch AI
Artificial Analysis Humanity's Last ExamKnowledge30.7%57.8——Artificial Analysis
Humanity's Last ExamKnowledge30.1%54.3——Center for AI Safety et al.
MCP AtlasAgentic29.5%30.5——OpenAI
FrontierMath-2025-02-28-PrivateMath27.9%56.7——Epoch AI
ToolathlonAgentic27.8%43.9——OpenAI
Artificial Analysis Intelligence IndexKnowledge23.5%51.7——Artificial Analysis
GDPval-AA normalizedAgentic16.2%48.0——Artificial Analysis
DeepPlanningAgentic14.4%———DeepPlanning authors
ResearchClawBenchAgentic14.0%———InternScience
Chess PuzzlesReasoning12.0%39.0——Epoch AI
ARC-AGI-2 (semi-private)Reasoning11.8%45.7——ARC Prize Foundation
APEX-Agents-AAAgentic11.5%50.2——Artificial Analysis / Mercor
JobBenchAgentic8.7%40.1——Yuetai Li et al.
FrontierMath-Tier-4-2025-07-01-PrivateMath4.2%48.0——Epoch AI
Critical Physics TasksReasoning3.1%44.9——Artificial Analysis

44 benchmarks count, from 50 of 65 results. A grey row does not count. Too few models took that benchmark.

Sources

BenchLM benchmark aggregationUsed with attribution; per-benchmark results credited to their original authorsOpenRouter, collected directlyNo licence statedEpoch AI, collected directlyCC BY — free to use and redistribute with attributionSWE-bench team, collected directlyNo licence stated. The repository publishes submission records for reproducibility and transparency and asks that SWE-bench be citedARC Prize Foundation, collected directlyNo licence stated. Their terms ask for written permission before commercial use

Same level, lower price

Qwen3.8-Flash-Next66.4 · Freedots3-note Preview65.5 · FreeApodex 1.1 Mini59.1 · Free

More from Moonshot AI

Kimi K372.0Kimi K2.660.9Kimi K2.7 Code59.5Kimi K2.5 (Reasoning)54.6Kimi K2.5 Thinking53.8