LLM Leaderboard
Compare quality, price, and speed across today's leading models
Top Klu Index100Claude Opus 5.5
Best value$2GPT-6.1 Terra
Fastest output930 tok/sK2 Horizon 375B A23B
Fastest first token0.5sCommand A+
Largest context1M tokensMuse Wonder
Most expensive$20 / 1MGPT-6 Astra
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
ExploreKlu Index
38 models
- All labs38
- Anthropic4
- Meta5
- OpenAI4
- Google3
- Mistral1
- NVIDIA3
- Thinking Machines2
- Other labs16
| # | Model | Klu Index | $ / 1M tokens | Tokens / s | First token | Context | Frontier |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 100 | $8.00 | 98 | 700s | 1M | Frontier |
| 2 | Claude Sonnet 5.5Anthropic | 97 | $4.00 | 140 | 455s | 1M | Frontier |
| 3 | Muse Wonder FullMuse Code subscription | 94 | Subscription | — | — | 1M | |
| 4 | GPT-6 AstraOpenAI | 93 | $20.00 | 71 | 269s | 1M | |
| 5 | Kimi K3.1 ProKimi | 93 | $4.00 | 41 | 2.8s | 1M | |
| 6 | Muse WonderMeta Model API and Muse Code | 92 | $8.00 | 208 | 31s | 1M | |
| 7 | Gemini 4 ArgonGoogle | 91 | $4.00 | 82 | 47s | 1M | |
| 8 | GPT-6.1 SolOpenAI | 91 | $4.00 | 81 | 278s | 1M | |
| 9 | Claude Fable 5.1Anthropic | 90 | $20.00 | 66 | 132s | 1M | |
| 10 | GPT-6.1 TerraOpenAI | 85 | $2.00 | 91 | 78s | 1M | Frontier |
| 11 | DeepSeek V4.1 Flash 1017DeepSeek | 84 | $0.53 | 173 | 1.1s | 1M | Frontier |
| 12 | Muse Spark 1.3Meta | 83 | $2.00 | 205 | 42s | 1M | |
| 13 | Grok 4.7xAI | 79 | $3.00 | 89 | 54s | 500k | |
| 14 | Step 5 PreviewStepFun | 78 | $1.43 | 183 | 1.2s | 1M | |
| 15 | GLM-5.3Z AI | 77 | $2.15 | 86 | 2.7s | 1M | |
| 16 | Qwen3.8 Max (0902)Alibaba | 77 | $3.00 | 42 | 3.1s | 984k | |
| 17 | Claude Haiku 5.5Anthropic | 76 | $0.20 | 241 | 311s | 1M | Frontier |
| 18 | Ling 3.1 Flash 1019InclusionAI | 75 | $0.45 | — | — | 1M | |
| 19 | MiniMax-M3.1-FlashMiniMax | 75 | $0.26 | 119 | 1.1s | 1M | |
| 20 | Muse Spark OpenMeta | 74 | $2.00 | 189 | 0.8s | 1M | |
| 21 | MiMo-V2.6-ProXiaomi | 73 | $0.54 | 57 | 3.2s | 1M | |
| 22 | Gemini 3.8 FlashGoogle | 69 | $1.50 | 133 | 27s | 1M | |
| 23 | K2 Horizon 375B A23BInstitute of Foundation Models | 69 | — | 930 | 2.3s | 512k | |
| 24 | MiMo-V2.6-FlashXiaomi | 69 | $0.18 | 58 | 3.3s | 1M | Frontier |
| 25 | Mistral Large 4 PreviewMistral API public preview | 69 | $2.07 | 114 | 1.6s | 512k | |
| 26 | DeepSeek V4 Pro 0813DeepSeek | 64 | $1.98 | 92 | 1.7s | 1M | |
| 27 | GPT-6 LunaOpenAI | 63 | $0.20 | 124 | 77s | 1M | |
| 28 | Beam 1.0Reflection | 62 | $0.60 | 371 | 0.6s | 1M | |
| 29 | Qwen3.8 27BAlibaba | 59 | $1.13 | 41 | 3.9s | 256k | |
| 30 | Nemotron 3.5 Super VLNVIDIA | 57 | $0.31 | 448 | 0.8s | 1M | |
| 31 | MiniMax-M3MiniMax | 55 | $0.53 | 109 | 1.0s | 1M | |
| 32 | Inkling SmallThinking Machines | 49 | $0.53 | 208 | 1.6s | 1M | |
| 33 | InklingThinking Machines | 47 | $1.76 | 194 | 1.8s | 1M | |
| 34 | Nemotron 3 Ultra 550B A55BNVIDIA | 47 | $1.13 | 199 | 1.3s | 256k | |
| 35 | Gemini 3.5 Flash-LiteGoogle | 45 | $0.85 | 461 | 6.6s | 1M | |
| 36 | Command A+Cohere | 32 | — | 152 | 0.5s | 192k | |
| 37 | Muse GlimmerMeta | 31 | $0.53 | 125 | 1.2s | 128k | |
| 38 | NVIDIA Nemotron 3 Nano 4BNVIDIA | 24 | — | — | — | 262k |
Muse Wonder Full is included in Muse Code Power Usage ($50/month) and Maximum Usage ($200/month). API token pricing and throughput are unavailable.
Mistral API public preview: Mistral launched October 6 with 1M context and a two-week offer: $0.68 input, $0.07 cached input, and $2.09 output per 1M tokens. Tables retain Artificial Analysis’s measured pricing and context.
Compare
- Claude Opus 5.5Anthropic100
- Claude Sonnet 5.5Anthropic97
- Muse Wonder FullMuse Code subscription94
- GPT-6 AstraOpenAI93
- Kimi K3.1 ProKimi93
- Muse WonderMeta Model API and Muse Code92
- Gemini 4 ArgonGoogle91
- GPT-6.1 SolOpenAI91
- Claude Fable 5.1Anthropic90
- GPT-6.1 TerraOpenAI85
- DeepSeek V4.1 Flash 1017DeepSeek84
- Muse Spark 1.3Meta83
- Grok 4.7xAI79
- Step 5 PreviewStepFun78
- GLM-5.3Z AI77
- Qwen3.8 Max (0902)Alibaba77
- Claude Haiku 5.5Anthropic76
- Ling 3.1 Flash 1019InclusionAI75
- MiniMax-M3.1-FlashMiniMax75
- Muse Spark OpenMeta74
- MiMo-V2.6-ProXiaomi73
- Gemini 3.8 FlashGoogle69
- K2 Horizon 375B A23BInstitute of Foundation Models69
- MiMo-V2.6-FlashXiaomi69
- Mistral Large 4 PreviewMistral API public preview69
- DeepSeek V4 Pro 0813DeepSeek64
- GPT-6 LunaOpenAI63
- Beam 1.0Reflection62
- Qwen3.8 27BAlibaba59
- Nemotron 3.5 Super VLNVIDIA57
- MiniMax-M3MiniMax55
- Inkling SmallThinking Machines49
- InklingThinking Machines47
- Nemotron 3 Ultra 550B A55BNVIDIA47
- Gemini 3.5 Flash-LiteGoogle45
- Command A+Cohere32
- Muse GlimmerMeta31
- NVIDIA Nemotron 3 Nano 4BNVIDIA24
- Claude Opus 5.5Anthropic100
- Claude Sonnet 5.5Anthropic97
- Muse Wonder FullMuse Code subscription94
- GPT-6 AstraOpenAI93
- Kimi K3.1 ProKimi93
- Muse WonderMeta Model API and Muse Code92
- Gemini 4 ArgonGoogle91
- GPT-6.1 SolOpenAI91
- Claude Fable 5.1Anthropic90
- GPT-6.1 TerraOpenAI85
- DeepSeek V4.1 Flash 1017DeepSeek84
- Muse Spark 1.3Meta83
- Grok 4.7xAI79
- Step 5 PreviewStepFun78
- GLM-5.3Z AI77
- Qwen3.8 Max (0902)Alibaba77
- Claude Haiku 5.5Anthropic76
- Ling 3.1 Flash 1019InclusionAI75
- MiniMax-M3.1-FlashMiniMax75
- Muse Spark OpenMeta74
- MiMo-V2.6-ProXiaomi73
- Gemini 3.8 FlashGoogle69
- K2 Horizon 375B A23BInstitute of Foundation Models69
- MiMo-V2.6-FlashXiaomi69
- Mistral Large 4 PreviewMistral API public preview69
- DeepSeek V4 Pro 0813DeepSeek64
- GPT-6 LunaOpenAI63
- Beam 1.0Reflection62
- Qwen3.8 27BAlibaba59
- Nemotron 3.5 Super VLNVIDIA57
- MiniMax-M3MiniMax55
- Inkling SmallThinking Machines49
- InklingThinking Machines47
- Nemotron 3 Ultra 550B A55BNVIDIA47
- Gemini 3.5 Flash-LiteGoogle45
- Command A+Cohere32
- Muse GlimmerMeta31
- NVIDIA Nemotron 3 Nano 4BNVIDIA24
100Klu Index97
$8 / 1M · 98 tok/sPrice and speed$4 / 1M · 140 tok/s
Frontier benchmark scores
| Model | AutomationBench | CritPt v2 Hard | Omniscience v1.1 | FrontierCode 2.0 | HLE-Diamond | T-B 4.0 | T-B Science | HealthBench Professional |
|---|---|---|---|---|---|---|---|---|
| Claude Opus 5.5 | 69.5 | 63.0 | 48.13 | 46.4 | 67.9 | 59.6 | 61.9 | 59.9 |
| Claude Sonnet 5.5 | 71.8 | 56.0 | 30.47 | 41.7 | 53.3 | 63.6 | 53.3 | 63.0 |
| Muse Wonder Full | 71.3 | 53.0 | 30.6 | 37.8 | 48.5 | 57.1 | 37.1 | Not reported |
| GPT-6 Astra | 68.5 | 71.5 | 44.57 | 42.4 | 63.8 | 59.6 | 63.3 | 66.0 |
| Kimi K3.1 Pro | 73.3 | 45.0 | 29.87 | 25.5 | 37.3 | 67.2 | 34.3 | Not reported |
| Muse Wonder | 68.1 | 48.0 | 25.1 | 27.9 | 39.0 | 52.0 | 25.7 | Not reported |
| Gemini 4 Argon | 77.5 | 52.5 | 46.1 | 25.7 | 56.7 | 57.1 | 52.9 | 55.1 |
| GPT-6.1 Sol | 66.6 | 68.5 | 42.37 | 40.6 | 58.8 | 56.1 | 58.1 | 65.1 |
| Claude Fable 5.1 | 59.4 | 60.5 | 46.27 | 35.7 | 60.1 | 55.1 | 43.3 | 57.8 |
| GPT-6.1 Terra | 60.2 | 64.5 | 29.57 | 42.3 | 47.3 | 55.6 | 57.6 | Not reported |
| DeepSeek V4.1 Flash 1017 | 68.9 | 24.0 | 0.9 | 28.5 | 25.3 | 33.8 | 12.9 | 50.7 |
| Muse Spark 1.3 | 57.9 | 46.0 | 23.83 | 23.1 | 33.7 | 33.3 | 11.0 | Not reported |
How the Klu Index works
The Klu Index combines accuracy, use-case evaluations, human preference, and speed into one score out of 100. The fastest way to compare models.
AccuracyArtificial Analysis Intelligence Index v5.0 across nine evaluations35%
EvaluationsLegal, finance, healthcare, engineering, economics, and strategy30%
PreferenceBlind pairwise judgments of finished work20%
PerformanceOutput speed and time to first token15%
Models by lab
Anthropic4 models
ModelKlu Index$ / 1M
Claude Opus 5.5100$8.00
Claude Sonnet 5.597$4.00
Claude Fable 5.190$20.00
Claude Haiku 5.576$0.20
Meta5 models
ModelKlu Index$ / 1M
Muse Wonder FullMuse Code subscription94—
Muse Wonder92$8.00
Muse Spark 1.383$2.00
Muse Spark Open74$2.00
Muse Glimmer31$0.53
OpenAI4 models
ModelKlu Index$ / 1M
GPT-6 Astra93$20.00
GPT-6.1 Sol91$4.00
GPT-6.1 Terra85$2.00
GPT-6 Luna63$0.20
Google3 models
ModelKlu Index$ / 1M
Gemini 4 Argon91$4.00
Gemini 3.8 Flash69$1.50
Gemini 3.5 Flash-Lite45$0.85
Mistral1 model
ModelKlu Index$ / 1M
Mistral Large 4 Preview69$2.07
NVIDIA3 models
ModelKlu Index$ / 1M
Nemotron 3.5 Super VL57$0.31
Nemotron 3 Ultra 550B A55B47$1.13
NVIDIA Nemotron 3 Nano 4B24—
Thinking Machines2 models
ModelKlu Index$ / 1M
Inkling Small49$0.53
Inkling47$1.76
Other labs16 models
ModelKlu Index$ / 1M
Kimi K3.1 Pro93$4.00
DeepSeek V4.1 Flash 101784$0.53
Grok 4.779$3.00
Step 5 Preview78$1.43
GLM-5.377$2.15
Qwen3.8 Max (0902)77$3.00
Ling 3.1 Flash 101975$0.45
MiniMax-M3.1-Flash75$0.26
MiMo-V2.6-Pro73$0.54
K2 Horizon 375B A23B69—
MiMo-V2.6-Flash69$0.18
DeepSeek V4 Pro 081364$1.98
Beam 1.062$0.60
Qwen3.8 27B59$1.13
MiniMax-M355$0.53
Command A+32—
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.