LLM Leaderboard

Compare quality, price, and speed across today's leading models

Updated · AA v5.0.1
Top Klu Index100Claude Fable 5.2
Best value$3Gemini 4 Hydrogen
Fastest output439 tok/sMuse Ember
Fastest first token0.7sNorth Mini Code
Largest context1.05M tokensGPT-6 Sol
Most expensive$20 / 1MClaude Fable 5.1
Klu Frontier IndexPrice per 1M tokens, log scale, cheaper to the right
FrontierAnthropicOpenAIGooglexAIMetaMistralOther
Frontier explorerKlu Index against cost per task, a point per reasoning effort
Explore
50100$0.005$0.020$0.10$0.50$2.00$10.0

Klu Index

67 models
#ModelKlu Index$ / 1M tokensTokens / sFirst tokenContextFrontier
1Claude Fable 5.2Anthropic100$16.00101271s1MFrontier
2Claude Opus 5.5Anthropic94$8.00114390s1MFrontier
3Gemini 4 CarbonGoogle93$4.006671s1MFrontier
4Gemini 4 HydrogenGoogle91$3.0018145s1MFrontier
5Claude Sonnet 5.5Anthropic90$4.00138266s1MNo
6Muse Wonder FullMuse Code subscription87Subscription——1MNo
7Claude Fable 5.1Anthropic84$20.0088204s1MNo
8Kimi K3.1 ProKimi84$4.00304.5s1MNo
9Gemini 4 ArgonGoogle83$4.006545s1MNo
10Muse WonderMeta Model API and Muse Code83$8.0011026s1MNo
11GPT-6 AstraOpenAI82$20.0036250s1MNo
12GPT-6.1 SolOpenAI82$4.0048201s1MNo
13Muse EmberMeta81$0.204395.0s1MFrontier
14Claude Opus 5Anthropic80$10.008155s1MNo
15DeepSeek V4.1 Flash 1017DeepSeek79$0.531621.3s1MNo
16GPT-6 SolOpenAI76$4.0011183s1.05MNo
17GPT-6.1 TerraOpenAI76$2.0068188s1MNo
18Muse Spark 1.3Meta76$2.0013229s1MNo
19Step 5 PreviewStepFun73$1.431330.9s1MNo
20Claude Haiku 5.5Anthropic72$0.20252175s1MNo
21Grok 4.7xAI72$3.006564s500kNo
22Qwen3.8 Max (0902)Alibaba72$3.00584.7s984kNo
23Tencent Hy4Tencent72$1.25913.0s1MNo
24GLM-5.3Z AI71$2.15753.1s1MNo
25Inkling 2 SmallThinking Machines71$0.48763.5s1.02MNo
26MiniMax-M3.1-FlashMiniMax71$0.261391.6s1MNo
27Ling 3.1 Flash 1019InclusionAI69$0.45——1MNo
28Grok 4.6xAI68$3.006433s500kNo
29Muse Spark OpenMeta68$2.001461.2s1MNo
30Qwen3.8-Flash-NextAlibaba68$0.23682.6s256kNo
31Agnes 2.5 Pro ThetaSapiens AI67$0.151702.1s1MFrontier
32MiMo-V2.6-ProXiaomi67$0.54564.7s1MNo
33Gemini 3.8 FlashGoogle65$1.5011317s1MNo
34GLM-5.3-FlashZ AI64$0.24473.7s1MNo
35Mistral Large 4Mistral API and open weights64$1.051083.4s1MNo
36MiMo-V2.6-FlashXiaomi62$0.18435.3s1MNo
37Mistral Large 4 PreviewRetired public preview62$1.05986.3s512kNo
38GPT-5.6 TerraOpenAI61$4.5099106s1MNo
39Beam 1.0Reflection58$0.602710.8s1MNo
40DeepSeek V4 Pro 0813DeepSeek58$1.98802.7s1MNo
41GPT-6 LunaOpenAI58$0.2010589s1MNo
42K2 Horizon 375B A23BInstitute of Foundation Models58—2504.8s512kNo
43GPT-5.6 LunaOpenAI55$0.4510676s1MNo
44Nanbeige4.3-33BNanbeige54———512kNo
45Qwen3.8 27BAlibaba54$1.13415.7s256kNo
46JT-4.1 Flash 236B A21BChina Mobile52———256kNo
47MiniMax-M3MiniMax52$0.531091.8s1MNo
48Nemotron 3.5 Super VLNVIDIA51$0.262211.1s1MNo
49Apodex 1.1Apodex46$0.981312.3s256kNo
50Solar Pro 4Upstage46$0.53842.3s512kNo
51Nemotron 3 Ultra 550B A55BNVIDIA45$1.131991.9s256kNo
52Inkling SmallThinking Machines44$0.53981.5s1MNo
53Kimi K3Kimi44$6.00427.4s1MNo
54Gemini 3.5 Flash-LiteGoogle43$0.852884.7s1MNo
55InklingThinking Machines41$1.76751.9s1MNo
56K-EXAONE 2.0LG AI Research34———256kNo
57Nemotron 3.5 LightningNVIDIA33$0.102590.9s1MFrontier
58Command A+Cohere32—1960.9s192kNo
59Gemini 3.1 Flash-LiteGoogle30$0.563054.1s1MNo
60Granite 4.2 3BIBM27———128kNo
61Muse GlimmerMeta27$0.53771.8s128kNo
62North Mini CodeCohere27—1770.7s256kNo
63Mistral Medium 3.5Mistral26$3.001467.2s256kNo
64Mistral Small 4Mistral26$0.261441.5s256kNo
65Granite 4.2 8BIBM22$0.11806.6s128kNo
66NVIDIA Nemotron 3 Nano 4BNVIDIA21———262kNo
67Gemma 4 31BGoogle20—301.0s256kNo

Muse Wonder Full is included in Muse Code Power Usage ($50/month) and Maximum Usage ($200/month). API token pricing and throughput are unavailable.

Retired public preview: The October 6 preview is retired. Historical preview measurements remain distinct from the final Mistral Large 4 release.

Compare

vs
100Klu Index94
76.4%AutomationBench69.5%
80.2%HLE-Diamond67.9%
61.6%T-B 4.059.6%
73.3%T-B Science61.9%
not reportedAgents' Last Exam (ALE)38.2%
not reportedArXivMath86%
not reportedBenchCAD0.962
not reportedChartography66.3%
not reportedCursorBench57.8%
not reportedFrontierMath95%
not reportedFrontierSWE v262.3%
not reportedGraphWalks90.6%
not reportedHarvey's Lab3.8%
not reportedMentalHealthBench52.4%
not reportedProgramBench18.5%
$16 / 1M · 101 tok/sPrice and speed$8 / 1M · 114 tok/s

Frontier benchmark scores

Primary-board scores for the top 12 models by Klu Index. Shading compares models within each benchmark only.
ModelAutomationBenchCritPt v2 HardOmniscience v1.1FrontierCode 2.0HLE-DiamondT-B 4.0T-B ScienceHealthBench Professional
Claude Fable 5.276.476.566.7749.180.261.673.368.1
Claude Opus 5.569.568.048.1346.467.959.661.959.9
Gemini 4 Carbon77.869.053.540.669.165.264.8Not reported
Gemini 4 Hydrogen76.166.046.2739.058.854.049.5Not reported
Claude Sonnet 5.571.859.030.4741.753.363.653.363.0
Muse Wonder Full71.356.530.637.848.557.137.1Not reported
Claude Fable 5.159.462.046.2735.760.155.143.357.8
Kimi K3.1 Pro73.349.529.8725.537.367.234.3Not reported
Gemini 4 Argon77.553.546.125.756.757.152.955.1
Muse Wonder68.151.025.127.939.052.025.7Not reported
GPT-6 Astra68.572.044.5742.463.859.663.366.0
GPT-6.1 Sol66.669.542.3740.658.856.158.165.1

How the Klu Index works

The Klu Index combines accuracy, use-case evaluations, human preference, and speed into one score out of 100. The fastest way to compare models.

AccuracyArtificial Analysis Intelligence Index v5.0.1 across nine evaluations35%
EvaluationsLegal, finance, healthcare, engineering, economics, and strategy30%
PreferenceBlind pairwise judgments of finished work20%
PerformanceOutput speed and time to first token15%

Models by lab

Anthropic6 models
ModelKlu Index$ / 1M
Claude Fable 5.2100$16.00
Claude Opus 5.594$8.00
Claude Sonnet 5.590$4.00
Claude Fable 5.184$20.00
Claude Opus 580$10.00
Claude Haiku 5.572$0.20
Google7 models
ModelKlu Index$ / 1M
Gemini 4 Carbon93$4.00
Gemini 4 Hydrogen91$3.00
Gemini 4 Argon83$4.00
Gemini 3.8 Flash65$1.50
Gemini 3.5 Flash-Lite43$0.85
Gemini 3.1 Flash-Lite30$0.56
Gemma 4 31B20—
Meta6 models
ModelKlu Index$ / 1M
Muse Wonder FullMuse Code subscription87—
Muse WonderMeta Model API and Muse Code83$8.00
Muse Ember81$0.20
Muse Spark 1.376$2.00
Muse Spark Open68$2.00
Muse Glimmer27$0.53
OpenAI7 models
ModelKlu Index$ / 1M
GPT-6 Astra82$20.00
GPT-6.1 Sol82$4.00
GPT-6 Sol76$4.00
GPT-6.1 Terra76$2.00
GPT-5.6 Terra61$4.50
GPT-6 Luna58$0.20
GPT-5.6 Luna55$0.45
Alibaba3 models
ModelKlu Index$ / 1M
Qwen3.8 Max (0902)72$3.00
Qwen3.8-Flash-Next68$0.23
Qwen3.8 27B54$1.13
Thinking Machines3 models
ModelKlu Index$ / 1M
Inkling 2 Small71$0.48
Inkling Small44$0.53
Inkling41$1.76
Mistral4 models
ModelKlu Index$ / 1M
Mistral Large 4Mistral API and open weights64$1.05
Mistral Large 4 PreviewRetired public preview62$1.05
Mistral Medium 3.526$3.00
Mistral Small 426$0.26
NVIDIA4 models
ModelKlu Index$ / 1M
Nemotron 3.5 Super VL51$0.26
Nemotron 3 Ultra 550B A55B45$1.13
Nemotron 3.5 Lightning33$0.10
NVIDIA Nemotron 3 Nano 4B21—
Other labs27 models
ModelKlu Index$ / 1M
Kimi K3.1 Pro84$4.00
DeepSeek V4.1 Flash 101779$0.53
Step 5 Preview73$1.43
Grok 4.772$3.00
Tencent Hy472$1.25
GLM-5.371$2.15
MiniMax-M3.1-Flash71$0.26
Ling 3.1 Flash 101969$0.45
Grok 4.668$3.00
Agnes 2.5 Pro Theta67$0.15
MiMo-V2.6-Pro67$0.54
GLM-5.3-Flash64$0.24
MiMo-V2.6-Flash62$0.18
Beam 1.058$0.60
DeepSeek V4 Pro 081358$1.98
K2 Horizon 375B A23B58—
Nanbeige4.3-33B54—
JT-4.1 Flash 236B A21B52—
MiniMax-M352$0.53
Apodex 1.146$0.98
Solar Pro 446$0.53
Kimi K344$6.00
K-EXAONE 2.034—
Command A+32—
Granite 4.2 3B27—
North Mini Code27—
Granite 4.2 8B22$0.11

About the leaderboard

What is the Klu Index?
One score out of 100 that blends public benchmark accuracy, use-case evaluations, blind preference judgments, and speed. Scores are scaled so the top model lands at 100.
How is speed measured?
Tokens per second is the median output speed on each provider's API. First token is the median time, in seconds, before the first token arrives.
How often is it updated?
Each time Klu refreshes the snapshot. The date at the top of the page shows when the current numbers were pulled.
Which model is best for me?
The one that wins on your own evals. Klu runs every model on your evals and shows which one comes out ahead for each use case.

The leaderboard is a starting point

The model at the top here isn't always the one that wins on your work. Klu runs every model on your evals and reruns them each time a new one ships.

Find your frontier