WhatsMyLLM 中文

Models in the fingerprint bank

The identify tool can recognise 24 models across 6 families, from 864 recorded answers. Each entry links to that model's own page: its favourite numbers, its full distribution, and how reliably it is recognised.

Bank version 2026.10.3. New models are added as they are collected; every change is listed on the changelog.

Capability and price

12 of the 24 models in the bank are ranked here: the ones whose provider still sells them and whose configuration the Artificial Analysis Intelligence Index (v4.3.2) still benchmarks in full. The figures were read on 2026-10-02–2026-10-17 and are those sources' claims, not measurements by this site. This tool identifies which model wrote a reply; it says nothing about how good that model is. The other 12 models keep their figures on their own pages.

#ModelProviderIndexInputOutputBlendedPrice #
1Claude Opus 5.5Anthropic584.0020.008.0010
2Claude Sonnet 5.5Anthropic562.0010.004.007
3Claude Fable 5.1Anthropic5310.0050.0020.0011
3GPT-6 AstraOpenAI5310.0050.0020.0011
5GPT-6.1 SolOpenAI522.0010.004.007
6Qwen3.8 Max 0902Alibaba452.006.003.006
7Claude Haiku 5.5Anthropic430.100.500.201
8GPT-5.6 TerraOpenAI422.0012.004.509
9Gemini 3.8 FlashGoogle410.753.751.503
10GPT-6 LunaOpenAI380.100.500.201
11DeepSeek V4 Pro 0813DeepSeek361.323.961.984
12Claude Haiku 4.5Anthropic151.005.002.005

Index is the composite score for the exact configuration that was scored; each model page names that configuration, and scores from different index versions are not comparable. Prices are US dollars per million tokens. Blended weights input three to one against output so that one number can order the table — a ranking aid, not a quote for your workload.

Catalogue

ModelFamilyCollected throughAnswersStatus
GPT-5.4
gpt-5.4
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
GPT-5.5
gpt-5.5
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
GPT-5.6 Sol
gpt-5.6-sol
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
GPT-5.6 Terra
gpt-5.6-terra
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
GPT-5.6 Luna
gpt-5.6-luna
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
GPT-6 Astra
gpt-6-astra
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
GPT-6 Sol
gpt-6-sol
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
GPT-6 Luna
gpt-6-luna
GPT API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Haiku 4.5
claude-haiku-4-5-20251001
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Sonnet 4.6
claude-sonnet-4-6
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Sonnet 5
claude-sonnet-5
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Opus 4.6
claude-opus-4-6
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Opus 4.7
claude-opus-4-7
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Opus 4.8
claude-opus-4-8
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Opus 5
claude-opus-5
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Opus 5.5
claude-opus-5-5
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Sonnet 5.5
claude-sonnet-5-5
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Haiku 5.5
claude-haiku-5-5
Claude API 36 Inherited from ModelTrace; serving identity not independently verified
Claude Fable 5.1
claude-fable-5-1@claude-code
Claude Claude Code 36 Collected by this project
Kimi K2.6
kimi-k2.6@cloudflare
Kimi Cloudflare Workers AI 36 Collected by this project
Gemini 3.8 Flash
gemini-3.8-flash@openrouter:google-vertex
Gemini OpenRouter 36 Collected by this project
DeepSeek V4 Pro 0813
deepseek-v4-pro-0813@openrouter:deepseek
DeepSeek OpenRouter 36 Collected by this project
Qwen3.8 Max 0902
qwen3.8-max-0902@openrouter:alibaba
Qwen OpenRouter 36 Collected by this project
GPT-6.1 Sol
gpt-6.1-sol
GPT API 36 Collected by this project

How reliably each model is recognised

ModelRunsClear match, rightClose callWeak matchClear match, wrongIf removed: clear match, wrong
GPT-5.41291200
GPT-5.51290300
GPT-5.6 Sol1292100
GPT-5.6 Terra12110100
GPT-5.6 Luna12120000
GPT-6 Astra12010209
GPT-6 Sol1253400
GPT-6 Luna1281300
Claude Haiku 4.512120000
Claude Sonnet 4.612110100
Claude Sonnet 512101101
Claude Opus 4.612120000
Claude Opus 4.712111006
Claude Opus 4.81255200
Claude Opus 512111005
Claude Opus 5.512120000
Claude Sonnet 5.512120000
Claude Haiku 5.512120000
Claude Fable 5.112100200
Kimi K2.61280400
Gemini 3.8 Flash1290300
DeepSeek V4 Pro 08131230900
Qwen3.8 Max 090212100200
GPT-6.1 Sol1209309

Each row repeats the measurement on that model's own page: held-out three-answer runs across all prompt environments, counted by the status the site attaches to the first-ranked model. The last column is the number to keep in mind if you are testing a version that is not listed here: with the model removed from the bank, how many of its own runs still named another model with a clear margin. The methodology page explains how both were measured.

What the status column means

18 entries were inherited from the ModelTrace project. Their answers were collected through an API relay whose serving identity was never independently confirmed, so they are marked as inherited rather than verified. That does not make them wrong: their fingerprints hold up in held-out testing exactly as the project's own do. It means the label on the tin is the one the relay reported.

A model collected by this project was answered by one independent session per challenge, with nothing carried over between them. The channel column says how the answers were obtained. A fingerprint tracks the model rather than the channel, so an entry collected through one channel still recognises the same model through another.