Data as of · 78 models, 80 questions · reviewed by LLM Verifications team · methodology

Providers · tested 27 Oct 2026

Compare AI labs by their best models

28 labs and 78 models, each tested through the same public API. Each lab is judged by its best model on every measure.

Most accurate modelLeading lab

Anthropic85.0%Claude Opus 5.5

Fastest modelLeading lab

NVIDIA715 tok/sNemotron 3.5 Super VL

Lowest latencyLeading lab

Mistral0.3 sMinistral 3 3B 2512

Cheapest per questionLeading lab

IBM$0.000051Granite 4.0 Micro

All labs

Ordered by each lab's best verification score. Select a lab for every model it makes.

4 models testedAnthropic85.0% best scoreBest: Claude Opus 5.5Fastest: Claude Haiku 5.5 · 192 tok/sCheapest: Claude Haiku 5.5 · $0.000486 models testedGoogle84.8% best scoreBest: Gemini 4 ArgonFastest: Gemini 3.5 Flash Lite · 409 tok/sCheapest: Gemma 4 26B A4B · $0.001113 models testedOpenAI82.5% best scoreBest: GPT-5.3-CodexFastest: o3 · 163 tok/sCheapest: gpt-oss-20b · $0.0000661 model testedSpaceXAI80.0% best scoreBest: Grok 4.7Fastest: Grok 4.7 · 77 tok/sCheapest: Grok 4.7 · $0.0255 models testedMeta78.8% best scoreBest: Muse WonderFastest: Muse Spark Open · 220 tok/sCheapest: Llama 4 Scout · $0.000372 models testedMoonshotAI77.5% best scoreBest: Kimi K3.1 ProFastest: Kimi K2.7 Code · 136 tok/sCheapest: Kimi K2.7 Code · $0.0069 models testedQwen75.0% best scoreBest: Qwen3.8 Max (0902)Fastest: Qwen3.5 397B A17B · 240 tok/sCheapest: Qwen3 Coder Next · $0.00173 models testedDeepSeek73.8% best scoreBest: DeepSeek V4.1 Flash 1017Fastest: DeepSeek V4 Flash Vision Exp · 170 tok/sCheapest: DeepSeek V4.1 Flash 1017 · $0.000912 models testedXiaomi68.8% best scoreBest: MiMo-V2.6-ProFastest: MiMo-V2.6-Flash · 77 tok/sCheapest: MiMo-V2.6-Flash · $0.000562 models testedThinking Machines67.5% best scoreBest: Inkling SmallFastest: Inkling Small · 233 tok/sCheapest: Inkling Small · $0.000472 models testedZ.ai67.5% best scoreBest: GLM 5.3 FlashFastest: GLM 5.3 · 154 tok/sCheapest: GLM 5.3 Flash · $0.000152 models testedMiniMax66.2% best scoreBest: MiniMax M3.1 FlashFastest: MiniMax M3.1 Flash · 131 tok/sCheapest: MiniMax M3.1 Flash · $0.000871 model testedStepFun63.8% best scoreBest: Step 5 PreviewFastest: Step 5 Preview · 93 tok/sCheapest: Step 5 Preview · $0.00473 models testedNVIDIA60.0% best scoreBest: Nemotron 3 UltraFastest: Nemotron 3.5 Super VL · 715 tok/sCheapest: Nemotron 3.5 Lightning · $0.000491 model testedTencent58.8% best scoreBest: Hy3Fastest: Hy3 · 96 tok/sCheapest: Hy3 · $0.00121 model testedInception58.2% best scoreBest: Mercury 2.5Fastest: Mercury 2.5 · 157 tok/sCheapest: Mercury 2.5 · $0.000116 models testedMistral50.0% best scoreBest: Mistral Large 4Fastest: Ministral 3 3B 2512 · 252 tok/sCheapest: Ministral 3 3B 2512 · $0.000191 model testedMeituan48.8% best scoreBest: LongCat 2.0Fastest: LongCat 2.0 · 67 tok/sCheapest: LongCat 2.0 · $0.00321 model testedCohere45.0% best scoreBest: Command AFastest: Command A · 7 tok/sCheapest: Command A · $0.0072 models testedUpstage45.0% best scoreBest: Solar Pro 4Fastest: Solar Mini 4 · 107 tok/sCheapest: Solar Mini 4 · $0.000532 models testedinclusionAI41.2% best scoreBest: Ling 3.1 FlashFastest: Ling 3.1 Flash · 7 tok/sCheapest: Ling 3.1 Flash · $0.000611 model testedMicrosoft41.2% best scoreBest: Phi 4Fastest: Phi 4 · 103 tok/sCheapest: Phi 4 · $0.000122 models testedIBM36.2% best scoreBest: Granite 4.2 8BFastest: Granite 4.2 8B · 60 tok/sCheapest: Granite 4.0 Micro · $0.0000511 model testedArcee AI28.8% best scoreBest: Trinity Large ThinkingFastest: Trinity Large Thinking · 329 tok/sCheapest: Trinity Large Thinking · $0.00251 model testedNous26.2% best scoreBest: Hermes 4 405BFastest: Hermes 4 405B · 41 tok/sCheapest: Hermes 4 405B · $0.012 models testedAmazon22.5% best scoreBest: Nova Micro 1.0Fastest: Nova Micro 1.0 · 180 tok/sCheapest: Nova Micro 1.0 · $0.0000871 model testedrekaai1.2% best scoreBest: Reka Flash 3Fastest: Reka Flash 3 · 101 tok/sCheapest: Reka Flash 3 · $0.00081 model testedLiquidAI—% best score

Accuracy

Each lab's highest verification score, with the model that achieved it.

Best verification score by lab

Highest-scoring model per lab · higher is better

  1. 1AnthropicClaude Opus 5.585.0
  2. 2GoogleGemini 4 Argon84.8
  3. 3OpenAIGPT-5.3-Codex82.5
  4. 4SpaceXAIGrok 4.780.0
  5. 5MetaMuse Wonder78.8
  6. 6MoonshotAIKimi K3.1 Pro77.5
  7. 7QwenQwen3.8 Max (0902)75.0
  8. 8DeepSeekDeepSeek V4.1 Flash 101773.8
  9. 9XiaomiMiMo-V2.6-Pro68.8
  10. 10Thinking MachinesInkling Small67.5
  11. 11Z.aiGLM 5.3 Flash67.5
  12. 12MiniMaxMiniMax M3.1 Flash66.2
  13. 13StepFunStep 5 Preview63.8
  14. 14NVIDIANemotron 3 Ultra60.0
  15. 15TencentHy358.8
  16. 16InceptionMercury 2.558.2
  17. 17MistralMistral Large 450.0
  18. 18MeituanLongCat 2.048.8
  19. 19CohereCommand A45.0
  20. 20UpstageSolar Pro 445.0
  21. 21inclusionAILing 3.1 Flash41.2
  22. 22MicrosoftPhi 441.2
  23. 23IBMGranite 4.2 8B36.2
  24. 24Arcee AITrinity Large Thinking28.8
  25. 25NousHermes 4 405B26.2
  26. 26AmazonNova Micro 1.022.5
  27. 27rekaaiReka Flash 31.2

Speed and latency

Each lab's fastest model, and its quickest to start responding.

Fastest model by lab

Median output tokens per second · higher is better

  1. 1NVIDIANemotron 3.5 Super VL715
  2. 2GoogleGemini 3.5 Flash Lite409
  3. 3Arcee AITrinity Large Thinking329
  4. 4MistralMinistral 3 3B 2512252
  5. 5QwenQwen3.5 397B A17B240
  6. 6Thinking MachinesInkling Small233
  7. 7MetaMuse Spark Open220
  8. 8AnthropicClaude Haiku 5.5192
  9. 9AmazonNova Micro 1.0180
  10. 10DeepSeekDeepSeek V4 Flash Vision Exp170
  11. 11OpenAIo3163
  12. 12InceptionMercury 2.5157
  13. 13Z.aiGLM 5.3154
  14. 14MoonshotAIKimi K2.7 Code136
  15. 15MiniMaxMiniMax M3.1 Flash131
  16. 16UpstageSolar Mini 4107
  17. 17MicrosoftPhi 4103
  18. 18rekaaiReka Flash 3101
  19. 19TencentHy396
  20. 20StepFunStep 5 Preview93
  21. 21XiaomiMiMo-V2.6-Flash77
  22. 22SpaceXAIGrok 4.777
  23. 23MeituanLongCat 2.067
  24. 24IBMGranite 4.2 8B60
  25. 25NousHermes 4 405B41
  26. 26CohereCommand A7
  27. 27inclusionAILing 3.1 Flash7

Lowest latency by lab

Median seconds to first token · lower is better

  1. 1MistralMinistral 3 3B 25120.3 s
  2. 2NousHermes 4 405B0.3 s
  3. 3MoonshotAIKimi K2.7 Code0.4 s
  4. 4XiaomiMiMo-V2.6-Flash0.4 s
  5. 5QwenQwen3.6 35B A3B0.4 s
  6. 6MicrosoftPhi 40.5 s
  7. 7AmazonNova Micro 1.00.5 s
  8. 8rekaaiReka Flash 30.5 s
  9. 9NVIDIANemotron 3.5 Lightning0.5 s
  10. 10MetaMuse Spark Open0.5 s
  11. 11Z.aiGLM 5.3 Flash0.5 s
  12. 12Thinking MachinesInkling Small0.5 s
  13. 13MiniMaxMiniMax M30.5 s
  14. 14OpenAIgpt-oss-120b0.5 s
  15. 15Arcee AITrinity Large Thinking0.6 s
  16. 16UpstageSolar Pro 40.6 s
  17. 17StepFunStep 5 Preview0.6 s
  18. 18IBMGranite 4.2 8B0.6 s
  19. 19GoogleGemma 4 31B0.6 s
  20. 20AnthropicClaude Haiku 5.50.7 s
  21. 21DeepSeekDeepSeek V4 Flash Vision Exp0.7 s
  22. 22inclusionAILing 3.1 Flash1.1 s
  23. 23MeituanLongCat 2.01.4 s
  24. 24TencentHy31.7 s
  25. 25CohereCommand A1.8 s
  26. 26SpaceXAIGrok 4.71.9 s
  27. 27InceptionMercury 2.52.3 s

Cost

Each lab's cheapest model per question as billed, and its largest context window.

Cheapest model by lab

Average billed USD per question · lower is better

  1. 1IBMGranite 4.0 Micro$0.000051
  2. 2OpenAIgpt-oss-20b$0.000066
  3. 3AmazonNova Micro 1.0$0.000087
  4. 4InceptionMercury 2.5$0.00011
  5. 5MicrosoftPhi 4$0.00012
  6. 6Z.aiGLM 5.3 Flash$0.00015
  7. 7MistralMinistral 3 3B 2512$0.00019
  8. 8MetaLlama 4 Scout$0.00037
  9. 9Thinking MachinesInkling Small$0.00047
  10. 10AnthropicClaude Haiku 5.5$0.00048
  11. 11NVIDIANemotron 3.5 Lightning$0.00049
  12. 12UpstageSolar Mini 4$0.00053
  13. 13XiaomiMiMo-V2.6-Flash$0.00056
  14. 14inclusionAILing 3.1 Flash$0.00061
  15. 15rekaaiReka Flash 3$0.0008
  16. 16MiniMaxMiniMax M3.1 Flash$0.00087
  17. 17DeepSeekDeepSeek V4.1 Flash 1017$0.00091
  18. 18GoogleGemma 4 26B A4B$0.0011
  19. 19TencentHy3$0.0012
  20. 20QwenQwen3 Coder Next$0.0017
  21. 21Arcee AITrinity Large Thinking$0.0025
  22. 22MeituanLongCat 2.0$0.0032
  23. 23StepFunStep 5 Preview$0.0047
  24. 24MoonshotAIKimi K2.7 Code$0.006
  25. 25CohereCommand A$0.007
  26. 26NousHermes 4 405B$0.01
  27. 27SpaceXAIGrok 4.7$0.025

Largest context window by lab

Maximum tokens per request · higher is better

  1. 1MetaLlama 4 Scout1.3M
  2. 2OpenAIGPT-6 Astra1.1M
  3. 3XiaomiMiMo-V2.6-Pro1.1M
  4. 4MeituanLongCat 2.01M
  5. 5GoogleGemini 4 Argon1M
  6. 6MoonshotAIKimi K3.1 Pro1M
  7. 7QwenQwen3.8 2.4T A95B1M
  8. 8DeepSeekDeepSeek V4.1 Flash 10171M
  9. 9Z.aiGLM 5.3 Flash1M
  10. 10MiniMaxMiniMax M3.1 Flash1M
  11. 11StepFunStep 5 Preview1M
  12. 12NVIDIANemotron 3.5 Super VL1M
  13. 13MistralMistral Large 41M
  14. 14AnthropicClaude Opus 5.51M
  15. 15AmazonNova Premier 1.01M
  16. 16Thinking MachinesInkling Small524k
  17. 17UpstageSolar Pro 4524k
  18. 18SpaceXAIGrok 4.7500k
  19. 19TencentHy3262k
  20. 20inclusionAILing 3.1 Flash262k
  21. 21Arcee AITrinity Large Thinking262k
  22. 22InceptionMercury 2.5260k
  23. 23CohereCommand A256k
  24. 24IBMGranite 4.2 8B131k
  25. 25NousHermes 4 405B131k
  26. 26LiquidAILFM2.5-2.6B128k
  27. 27rekaaiReka Flash 366k
  28. 28MicrosoftPhi 416k

Comparison table

LabModelsBest scoreFastestLowest latencyCheapest / questionLargest context
Anthropic485.0%Claude Opus 5.5192 tok/sClaude Haiku 5.50.7 sClaude Haiku 5.5$0.00048Claude Haiku 5.51MClaude Opus 5.5
Google684.8%Gemini 4 Argon409 tok/sGemini 3.5 Flash Lite0.6 sGemma 4 31B$0.0011Gemma 4 26B A4B1MGemini 4 Argon
OpenAI1382.5%GPT-5.3-Codex163 tok/so30.5 sgpt-oss-120b$0.000066gpt-oss-20b1.1MGPT-6 Astra
SpaceXAI180.0%Grok 4.777 tok/sGrok 4.71.9 sGrok 4.7$0.025Grok 4.7500kGrok 4.7
Meta578.8%Muse Wonder220 tok/sMuse Spark Open0.5 sMuse Spark Open$0.00037Llama 4 Scout1.3MLlama 4 Scout
MoonshotAI277.5%Kimi K3.1 Pro136 tok/sKimi K2.7 Code0.4 sKimi K2.7 Code$0.006Kimi K2.7 Code1MKimi K3.1 Pro
Qwen975.0%Qwen3.8 Max (0902)240 tok/sQwen3.5 397B A17B0.4 sQwen3.6 35B A3B$0.0017Qwen3 Coder Next1MQwen3.8 2.4T A95B
DeepSeek373.8%DeepSeek V4.1 Flash 1017170 tok/sDeepSeek V4 Flash Vision Exp0.7 sDeepSeek V4 Flash Vision Exp$0.00091DeepSeek V4.1 Flash 10171MDeepSeek V4.1 Flash 1017
Xiaomi268.8%MiMo-V2.6-Pro77 tok/sMiMo-V2.6-Flash0.4 sMiMo-V2.6-Flash$0.00056MiMo-V2.6-Flash1.1MMiMo-V2.6-Pro
Thinking Machines267.5%Inkling Small233 tok/sInkling Small0.5 sInkling Small$0.00047Inkling Small524kInkling Small
Z.ai267.5%GLM 5.3 Flash154 tok/sGLM 5.30.5 sGLM 5.3 Flash$0.00015GLM 5.3 Flash1MGLM 5.3 Flash
MiniMax266.2%MiniMax M3.1 Flash131 tok/sMiniMax M3.1 Flash0.5 sMiniMax M3$0.00087MiniMax M3.1 Flash1MMiniMax M3.1 Flash
StepFun163.8%Step 5 Preview93 tok/sStep 5 Preview0.6 sStep 5 Preview$0.0047Step 5 Preview1MStep 5 Preview
NVIDIA360.0%Nemotron 3 Ultra715 tok/sNemotron 3.5 Super VL0.5 sNemotron 3.5 Lightning$0.00049Nemotron 3.5 Lightning1MNemotron 3.5 Super VL
Tencent158.8%Hy396 tok/sHy31.7 sHy3$0.0012Hy3262kHy3
Inception158.2%Mercury 2.5157 tok/sMercury 2.52.3 sMercury 2.5$0.00011Mercury 2.5260kMercury 2.5
Mistral650.0%Mistral Large 4252 tok/sMinistral 3 3B 25120.3 sMinistral 3 3B 2512$0.00019Ministral 3 3B 25121MMistral Large 4
Meituan148.8%LongCat 2.067 tok/sLongCat 2.01.4 sLongCat 2.0$0.0032LongCat 2.01MLongCat 2.0
Cohere145.0%Command A7 tok/sCommand A1.8 sCommand A$0.007Command A256kCommand A
Upstage245.0%Solar Pro 4107 tok/sSolar Mini 40.6 sSolar Pro 4$0.00053Solar Mini 4524kSolar Pro 4
inclusionAI241.2%Ling 3.1 Flash7 tok/sLing 3.1 Flash1.1 sLing 3.1 Flash$0.00061Ling 3.1 Flash262kLing 3.1 Flash
Microsoft141.2%Phi 4103 tok/sPhi 40.5 sPhi 4$0.00012Phi 416kPhi 4
IBM236.2%Granite 4.2 8B60 tok/sGranite 4.2 8B0.6 sGranite 4.2 8B$0.000051Granite 4.0 Micro131kGranite 4.2 8B
Arcee AI128.8%Trinity Large Thinking329 tok/sTrinity Large Thinking0.6 sTrinity Large Thinking$0.0025Trinity Large Thinking262kTrinity Large Thinking
Nous126.2%Hermes 4 405B41 tok/sHermes 4 405B0.3 sHermes 4 405B$0.01Hermes 4 405B131kHermes 4 405B
Amazon222.5%Nova Micro 1.0180 tok/sNova Micro 1.00.5 sNova Micro 1.0$0.000087Nova Micro 1.01MNova Premier 1.0
rekaai11.2%Reka Flash 3101 tok/sReka Flash 30.5 sReka Flash 3$0.0008Reka Flash 366kReka Flash 3
LiquidAI1————128kLFM2.5-2.6B

Frequently asked questions

Which AI lab makes the most accurate model?
Anthropic: Claude Opus 5.5 has the highest verification score we measured in our 2026-10-27 run, at 85.0%.
Which AI lab has the fastest model?
NVIDIA: Nemotron 3.5 Super VL had the highest median output speed in our 2026-10-27 run, at 715 tokens per second.
Which AI lab has the cheapest model?
IBM: Granite 4.0 Micro had the lowest billed cost per question in our 2026-10-27 run, at $0.000051.
Which AI lab has the most models on this site?
OpenAI, with 13 models tested: GPT-5.3-Codex, GPT-6 Astra, GPT-6.1 Terra Pro, GPT-6.1 Sol, GPT-5.6 Terra, GPT-6.1 Terra, GPT-6 Sol, o3, GPT-5.4, GPT-6 Luna, gpt-oss-120b, gpt-oss-20b, GPT-5.5 Pro.
How many AI labs do you compare?
28 labs, 78 models in total.