Skip to content Chat
Console

Model speed

How fast each model answers measured on the OceanToken API

73 text models from 33 makers. Measured on October 29, 2026, median of 3 runs.

Fastest output

729 tokens/s

MiMo-V2.6-Pro-UltraSpeed

  1. Mercury 2.5618 tokens/s
  2. Inkling Small455 tokens/s

Quickest first token

1.22 s

Seed-2.0-Mini

  1. Laguna S 2.11.26 s
  2. Granite 4.2 8B1.28 s

Typical model

82 tokens/s

2.40 s to first token

Every model on one chart

Up and to the left is better. Models beyond the axis sit at its edge.

01002003000246810MiMo-V2.6-Pro-UltraSpeed: 729 tokens/s, 2.37 sMercury 2.5: 618 tokens/s, 5.50 sInkling Small: 455 tokens/s, 1.62 sNemotron 3 Super: 372 tokens/s, 3.94 sLing 3.0 Flash: 293 tokens/s, 1.62 sGemini 3.5 Flash Lite: 244 tokens/s, 1.88 sNova Micro 1.0: 239 tokens/s, 2.09 sMuse Glimmer 30B: 216 tokens/s, 1.46 sGLM 5.3 FlashX: 195 tokens/s, 2.58 sNemotron 3 Ultra: 190 tokens/s, 1.59 sNova 2 Lite: 190 tokens/s, 2.40 sNova Pro 1.0: 166 tokens/s, 2.26 sDeepSeek V4.1 Flash: 165 tokens/s, 2.95 sCommand A+: 156 tokens/s, 1.80 sGLM 5.3 Prime: 150 tokens/s, 1.46 sGLM 5.3: 148 tokens/s, 1.64 sKimi K3: 148 tokens/s, 1.75 sGPT-6.1 Terra: 148 tokens/s, 5.47 sGemini 3.8 Flash: 145 tokens/s, 17.08 sMiniMax M3: 144 tokens/s, 3.30 sMistral Small 4: 137 tokens/s, 1.47 sPerceptron Mk1.5: 137 tokens/s, 1.97 sGPT-6 Luna: 135 tokens/s, 9.42 sSeed-2.0-Mini: 124 tokens/s, 1.22 sNemotron 3.5 Lightning: 115 tokens/s, 1.96 sInkling: 113 tokens/s, 1.47 sHy3: 111 tokens/s, 1.77 sPhi 4: 103 tokens/s, 1.51 sClaude Sonnet 5.5: 103 tokens/s, 14.28 sSolar Mini 4: 101 tokens/s, 1.87 sLaguna S 2.1: 97 tokens/s, 1.26 sLongCat 2.0: 97 tokens/s, 2.56 sReka Flash 3: 95 tokens/s, 4.24 sQwen3.8 Max Prime: 93 tokens/s, 2.24 sMistral Medium 3.5: 90 tokens/s, 2.19 sSolar Pro 4: 85 tokens/s, 2.05 sStep 3.7 Flash: 82 tokens/s, 6.30 sClaude Haiku 4.5: 79 tokens/s, 2.72 sQwen3.8 Flash: 75 tokens/s, 1.97 sMiMo-V2.6-Pro: 75 tokens/s, 2.47 sDeepSeek V4 Pro: 75 tokens/s, 3.05 sAion 3.5 Mini: 73 tokens/s, 3.91 sCommand R7B (12-2024): 71 tokens/s, 1.49 sMiniMax M2.5: 71 tokens/s, 3.43 sClaude Opus 5.5: 69 tokens/s, 15.86 sEmber-1: 66 tokens/s, 2.93 sPalmyra X5: 65 tokens/s, 2.34 sGemini 4 Argon: 65 tokens/s, 9.76 sAion 3.5: 62 tokens/s, 2.87 sStep 3.5 Flash: 59 tokens/s, 1.98 sMiniMax M2.7: 59 tokens/s, 3.44 sKimi K2.6: 58 tokens/s, 2.11 sGranite 4.2 8B: 55 tokens/s, 1.28 sAion-3.0: 53 tokens/s, 2.79 sDeepSeek V3.2: 53 tokens/s, 2.91 sHy4 preview: 53 tokens/s, 3.81 sQwen3.7 Plus: 52 tokens/s, 2.59 sHunyuan A13B Instruct: 51 tokens/s, 2.35 sMiMo-V2.6-Flash: 46 tokens/s, 6.03 sHermes 4 405B: 45 tokens/s, 2.01 sKimi K2.7 Code: 42 tokens/s, 2.43 sMuse Spark 1.3: 36 tokens/s, 2.82 sHermes 3 70B Instruct: 36 tokens/s, 3.40 sGrok 4.7: 35 tokens/s, 2.38 sGPT-6.1 Sol: 35 tokens/s, 9.42 sLlama 3.3 70B Instruct: 33 tokens/s, 3.18 sGrok 4.3: 28 tokens/s, 2.91 sGrok Build 0.1: 25 tokens/s, 2.82 sCommand A: 19 tokens/s, 2.26 sTernary Bonsai 2 27B: 19 tokens/s, 3.72 sSeed 2.1 Turbo: 14 tokens/s, 2.00 sLlama 4 Maverick: 13 tokens/s, 2.42 sSeed-2.0-Lite: 12 tokens/s, 1.94 sTime to first token (s)Output speed (tokens/s)

All 73 models

ModelMaker
MiMo-V2.6-Pro-UltraSpeedXiaomi729 tokens/s2.37 s
Mercury 2.5Inception618 tokens/s5.50 s
Inkling SmallThinking Machines455 tokens/s1.62 s
Nemotron 3 SuperNVIDIA372 tokens/s3.94 s
Ling 3.0 FlashinclusionAI293 tokens/s1.62 s
Gemini 3.5 Flash LiteGoogle244 tokens/s1.88 s
Nova Micro 1.0Amazon239 tokens/s2.09 s
Muse Glimmer 30BMeta216 tokens/s1.46 s
GLM 5.3 FlashXZ.ai195 tokens/s2.58 s
Nemotron 3 UltraNVIDIA190 tokens/s1.59 s
Nova 2 LiteAmazon190 tokens/s2.40 s
Nova Pro 1.0Amazon166 tokens/s2.26 s
DeepSeek V4.1 FlashDeepSeek165 tokens/s2.95 s
Command A+Cohere156 tokens/s1.80 s
GLM 5.3 PrimeZ.ai150 tokens/s1.46 s
GLM 5.3Z.ai148 tokens/s1.64 s
Kimi K3MoonshotAI148 tokens/s1.75 s
GPT-6.1 TerraOpenAI148 tokens/s5.47 s
Gemini 3.8 FlashGoogle145 tokens/s17.08 s
MiniMax M3MiniMax144 tokens/s3.30 s
Mistral Small 4Mistral137 tokens/s1.47 s
Perceptron Mk1.5Perceptron137 tokens/s1.97 s
GPT-6 LunaOpenAI135 tokens/s9.42 s
Seed-2.0-MiniByteDance Seed124 tokens/s1.22 s
Nemotron 3.5 LightningNVIDIA115 tokens/s1.96 s
InklingThinking Machines113 tokens/s1.47 s
Hy3Tencent111 tokens/s1.77 s
Phi 4Microsoft AI103 tokens/s1.51 s
Claude Sonnet 5.5Anthropic103 tokens/s14.28 s
Solar Mini 4Upstage101 tokens/s1.87 s
Laguna S 2.1Poolside97 tokens/s1.26 s
LongCat 2.0Meituan97 tokens/s2.56 s
Reka Flash 3Rekaai95 tokens/s4.24 s
Qwen3.8 Max PrimeQwen93 tokens/s2.24 s
Mistral Medium 3.5Mistral90 tokens/s2.19 s
Solar Pro 4Upstage85 tokens/s2.05 s
Step 3.7 FlashStepFun82 tokens/s6.30 s
Claude Haiku 4.5Anthropic79 tokens/s2.72 s
Qwen3.8 FlashQwen75 tokens/s1.97 s
MiMo-V2.6-ProXiaomi75 tokens/s2.47 s
DeepSeek V4 ProDeepSeek75 tokens/s3.05 s
Aion 3.5 MiniAionLabs73 tokens/s3.91 s
Command R7B (12-2024)Cohere71 tokens/s1.49 s
MiniMax M2.5MiniMax71 tokens/s3.43 s
Claude Opus 5.5Anthropic69 tokens/s15.86 s
Ember-1Fireworks66 tokens/s2.93 s
Palmyra X5Writer65 tokens/s2.34 s
Gemini 4 ArgonGoogle65 tokens/s9.76 s
Aion 3.5AionLabs62 tokens/s2.87 s
Step 3.5 FlashStepFun59 tokens/s1.98 s
MiniMax M2.7MiniMax59 tokens/s3.44 s
Kimi K2.6MoonshotAI58 tokens/s2.11 s
Granite 4.2 8BIBM55 tokens/s1.28 s
Aion-3.0AionLabs53 tokens/s2.79 s
DeepSeek V3.2DeepSeek53 tokens/s2.91 s
Hy4 previewTencent53 tokens/s3.81 s
Qwen3.7 PlusQwen52 tokens/s2.59 s
Hunyuan A13B InstructTencent51 tokens/s2.35 s
MiMo-V2.6-FlashXiaomi46 tokens/s6.03 s
Hermes 4 405BNous45 tokens/s2.01 s
Kimi K2.7 CodeMoonshotAI42 tokens/s2.43 s
Muse Spark 1.3Meta36 tokens/s2.82 s
Hermes 3 70B InstructNous36 tokens/s3.40 s
Grok 4.7SpaceXAI35 tokens/s2.38 s
GPT-6.1 SolOpenAI35 tokens/s9.42 s
Llama 3.3 70B InstructMeta33 tokens/s3.18 s
Grok 4.3SpaceXAI28 tokens/s2.91 s
Grok Build 0.1SpaceXAI25 tokens/s2.82 s
Command ACohere19 tokens/s2.26 s
Ternary Bonsai 2 27BPrismML19 tokens/s3.72 s
Seed 2.1 TurboByteDance Seed14 tokens/s2.00 s
Llama 4 MaverickMeta13 tokens/s2.42 s
Seed-2.0-LiteByteDance Seed12 tokens/s1.94 s

How we measure

We send each model the same streaming request through the OceanToken API 3 times and take the median. Time to first token runs from sending the request to the first token back; output speed is the tokens returned per second after that. Speed changes with the time of day, the length of the request and the load, so your results may differ.

Try any of these models with one API key

Get API Key See all models and prices