TokenCounts logo
TokenCountsMulti-provider token counts

Monthly Tokenization Benchmark Report

Tokenization benchmarks measure subword splitting efficiency. Across standard prompts, OpenAI GPT-4o has the highest compression ratio (average 4.30 chars/token on code), followed by Gemini 3.5 (average 4.20 chars/token), while Claude Sonnet 4.6 (average 3.94 chars/token) splits words into smaller token segments.

The Great Gatsby (Chapter 1) - Prose

Standard English literary prose containing narrative, descriptions, and dialogue.

Payload details: 5,737 words · 32,751 characters

Model NameToken CountExpansion Ratio (Chars/Token)
Claude Opus 4.710,8143.03 chars/token
Claude Opus 4.810,8093.03 chars/token
Claude Fable 58,0364.08 chars/token
Claude Opus 4.68,0364.08 chars/token
Claude Opus 4.58,0364.08 chars/token
Claude Sonnet 4.68,0364.08 chars/token
Claude Sonnet 4.58,0364.08 chars/token
Claude Haiku 4.58,0364.08 chars/token
Gemini 3.5 Flash7,5034.37 chars/token
Gemini 3.1 Pro7,5034.37 chars/token
Gemini 3 Flash7,5034.37 chars/token
Gemini 3.1 Flash-Lite7,5034.37 chars/token
Gemini 2.5 Pro7,5034.37 chars/token
Gemini 2.5 Flash7,5034.37 chars/token
Gemini 2.5 Flash-Lite7,5034.37 chars/token
Gemma 4 E4B7,5004.37 chars/token
Gemma 4 26B A4B7,5004.37 chars/token
Gemma 4 31B7,5004.37 chars/token
Qwen 3.5 9B7,3884.43 chars/token
Qwen 3.6 27B7,3884.43 chars/token
Mistral Small 4 (119B)7,3604.45 chars/token
Llama 4 Scout (17B)7,3024.49 chars/token
Llama 4 Maverick (17B)7,3024.49 chars/token
GPT-47,2634.51 chars/token
GPT-3.5 Turbo7,2634.51 chars/token
Kimi K2.67,2554.51 chars/token
DeepSeek V4 Flash7,2464.52 chars/token
DeepSeek V4 Pro7,2464.52 chars/token
GPT-5.57,1834.56 chars/token
GPT-5.5 Pro7,1834.56 chars/token
GPT-5.4 mini7,1834.56 chars/token
GPT-5.4 nano7,1834.56 chars/token
GPT-4o7,1834.56 chars/token
GPT-4o mini7,1834.56 chars/token
o37,1834.56 chars/token
o4-mini7,1834.56 chars/token

Standard Mutual NDA Agreement - Legal Text

Complex legal agreement with dense wording, bullet points, and references.

Payload details: 2,400 words · 16,415 characters

Model NameToken CountExpansion Ratio (Chars/Token)
Claude Opus 4.83,4204.8 chars/token
Claude Fable 53,4204.8 chars/token
Claude Opus 4.73,4204.8 chars/token
Claude Opus 4.63,4204.8 chars/token
Claude Opus 4.53,4204.8 chars/token
Claude Sonnet 4.63,4204.8 chars/token
Claude Sonnet 4.53,4204.8 chars/token
Claude Haiku 4.53,4204.8 chars/token
Mistral Small 4 (119B)3,2195.1 chars/token
Llama 4 Scout (17B)3,1585.2 chars/token
Llama 4 Maverick (17B)3,1585.2 chars/token
DeepSeek V4 Flash3,1305.24 chars/token
DeepSeek V4 Pro3,1305.24 chars/token
GPT-43,1135.27 chars/token
GPT-3.5 Turbo3,1135.27 chars/token
Qwen 3.5 9B3,1065.28 chars/token
Qwen 3.6 27B3,1065.28 chars/token
Kimi K2.63,0895.31 chars/token
GPT-5.53,0835.32 chars/token
GPT-5.5 Pro3,0835.32 chars/token
GPT-5.4 mini3,0835.32 chars/token
GPT-5.4 nano3,0835.32 chars/token
GPT-4o3,0835.32 chars/token
GPT-4o mini3,0835.32 chars/token
o33,0835.32 chars/token
o4-mini3,0835.32 chars/token
Gemini 3.5 Flash3,0705.35 chars/token
Gemini 3.1 Pro3,0705.35 chars/token
Gemini 3 Flash3,0705.35 chars/token
Gemini 3.1 Flash-Lite3,0705.35 chars/token
Gemini 2.5 Pro3,0705.35 chars/token
Gemini 2.5 Flash3,0705.35 chars/token
Gemini 2.5 Flash-Lite3,0705.35 chars/token
Gemma 4 E4B3,0705.35 chars/token
Gemma 4 26B A4B3,0705.35 chars/token
Gemma 4 31B3,0705.35 chars/token

Main App Controller (React/ES6)

JavaScript React component code containing import syntax, brackets, variables, and logic statements.

Payload details: 2,462 words · 25,000 characters

Model NameToken CountExpansion Ratio (Chars/Token)
Claude Opus 4.87,5143.33 chars/token
Claude Fable 57,5143.33 chars/token
Claude Opus 4.77,5143.33 chars/token
Claude Opus 4.67,5143.33 chars/token
Claude Opus 4.57,5143.33 chars/token
Claude Sonnet 4.67,5143.33 chars/token
Claude Sonnet 4.57,5143.33 chars/token
Claude Haiku 4.57,5143.33 chars/token
Gemini 3.5 Flash7,0083.57 chars/token
Gemini 3.1 Pro7,0083.57 chars/token
Gemini 3 Flash7,0083.57 chars/token
Gemini 3.1 Flash-Lite7,0083.57 chars/token
Gemini 2.5 Pro7,0083.57 chars/token
Gemini 2.5 Flash7,0083.57 chars/token
Gemini 2.5 Flash-Lite7,0083.57 chars/token
Gemma 4 E4B7,0083.57 chars/token
Gemma 4 26B A4B7,0083.57 chars/token
Gemma 4 31B7,0083.57 chars/token
Qwen 3.5 9B6,3843.92 chars/token
Qwen 3.6 27B6,3843.92 chars/token
Mistral Small 4 (119B)6,1234.08 chars/token
DeepSeek V4 Flash6,0624.12 chars/token
DeepSeek V4 Pro6,0624.12 chars/token
Kimi K2.65,9194.22 chars/token
GPT-5.55,8974.24 chars/token
GPT-5.5 Pro5,8974.24 chars/token
GPT-5.4 mini5,8974.24 chars/token
GPT-5.4 nano5,8974.24 chars/token
GPT-4o5,8974.24 chars/token
GPT-4o mini5,8974.24 chars/token
o35,8974.24 chars/token
o4-mini5,8974.24 chars/token
Llama 4 Scout (17B)5,8904.24 chars/token
Llama 4 Maverick (17B)5,8904.24 chars/token
GPT-45,6174.45 chars/token
GPT-3.5 Turbo5,6174.45 chars/token

Structured JSON Database Dump

Large structured JSON file with repeated keys, brackets, indentation, and numbers.

Payload details: 15,347 words · 180,771 characters

Model NameToken CountExpansion Ratio (Chars/Token)
Gemini 3.5 Flash62,1492.91 chars/token
Gemini 3.1 Pro62,1492.91 chars/token
Gemini 3 Flash62,1492.91 chars/token
Gemini 3.1 Flash-Lite62,1492.91 chars/token
Gemini 2.5 Pro62,1492.91 chars/token
Gemini 2.5 Flash62,1492.91 chars/token
Gemini 2.5 Flash-Lite62,1492.91 chars/token
Gemma 4 E4B62,1492.91 chars/token
Gemma 4 26B A4B62,1492.91 chars/token
Gemma 4 31B62,1492.91 chars/token
Qwen 3.5 9B60,5992.98 chars/token
Qwen 3.6 27B60,5992.98 chars/token
Claude Opus 4.858,4453.09 chars/token
Claude Fable 558,4453.09 chars/token
Claude Opus 4.758,4453.09 chars/token
Claude Opus 4.658,4453.09 chars/token
Claude Opus 4.558,4453.09 chars/token
Claude Sonnet 4.658,4453.09 chars/token
Claude Sonnet 4.558,4453.09 chars/token
Claude Haiku 4.558,4453.09 chars/token
Mistral Small 4 (119B)56,4133.2 chars/token
DeepSeek V4 Flash51,3073.52 chars/token
DeepSeek V4 Pro51,3073.52 chars/token
GPT-5.550,5393.58 chars/token
GPT-5.5 Pro50,5393.58 chars/token
GPT-5.4 mini50,5393.58 chars/token
GPT-5.4 nano50,5393.58 chars/token
GPT-4o50,5393.58 chars/token
GPT-4o mini50,5393.58 chars/token
o350,5393.58 chars/token
o4-mini50,5393.58 chars/token
Kimi K2.650,5323.58 chars/token
Llama 4 Scout (17B)50,3783.59 chars/token
Llama 4 Maverick (17B)50,3783.59 chars/token
GPT-449,7643.63 chars/token
GPT-3.5 Turbo49,7643.63 chars/token

10-Page Research PDF with Images & Tables

Scientific paper PDF containing layout styling, 2 charts (images), and multiple data tables.

Payload details: 4,500 words · 32,622 characters

Model NameToken CountExpansion Ratio (Chars/Token)
Claude Opus 4.87,6264.28 chars/token
Claude Fable 57,6264.28 chars/token
Claude Opus 4.77,6264.28 chars/token
Claude Opus 4.67,6264.28 chars/token
Claude Opus 4.57,6264.28 chars/token
Claude Sonnet 4.67,6264.28 chars/token
Claude Sonnet 4.57,6264.28 chars/token
Claude Haiku 4.57,6264.28 chars/token
Mistral Small 4 (119B)6,9424.7 chars/token
Gemini 3.5 Flash6,9244.71 chars/token
Gemini 3.1 Pro6,9244.71 chars/token
Gemini 3 Flash6,9244.71 chars/token
Gemini 3.1 Flash-Lite6,9244.71 chars/token
Gemini 2.5 Pro6,9244.71 chars/token
Gemini 2.5 Flash6,9244.71 chars/token
Gemini 2.5 Flash-Lite6,9244.71 chars/token
Gemma 4 E4B6,9244.71 chars/token
Gemma 4 26B A4B6,9244.71 chars/token
Gemma 4 31B6,9244.71 chars/token
Qwen 3.5 9B6,8734.75 chars/token
Qwen 3.6 27B6,8734.75 chars/token
GPT-46,7434.84 chars/token
GPT-3.5 Turbo6,7434.84 chars/token
DeepSeek V4 Flash6,7394.84 chars/token
DeepSeek V4 Pro6,7394.84 chars/token
Kimi K2.66,6884.88 chars/token
Llama 4 Scout (17B)6,6714.89 chars/token
Llama 4 Maverick (17B)6,6714.89 chars/token
GPT-5.56,6574.9 chars/token
GPT-5.5 Pro6,6574.9 chars/token
GPT-5.4 mini6,6574.9 chars/token
GPT-5.4 nano6,6574.9 chars/token
GPT-4o6,6574.9 chars/token
GPT-4o mini6,6574.9 chars/token
o36,6574.9 chars/token
o4-mini6,6574.9 chars/token