YFarmX logoYFarmXIntelligence for an accelerating worldBriefings
YFarmX logoYFarmX
YFarmX
AI Desk · AI Fingerprinting

Unknowns

Two kinds of blank, kept apart. A stealth model hides its maker: nobody says whose it is, and measurement is the only way to find out. An unlabelled model has a maker on the tin but no tokenizer family on its catalogue label, and measurement fills that blank in.

The stealth lab

Models whose maker is hidden on purpose, ranked by evidence. These are the true unknowns, and each one gets measured while it hides.

Union Alpha

stealth/union-alpha · Circuit and ChiselConfirmed by its maker
Confirmed family
Pareto (multi-model blend)
Confirmed as
Pareto 26.9, built by Circuit and Chisel
Confirmed by The Unbiased Co
17 September 2026
Entered catalogue
16 September 2026
Last tested
17 September 2026
Family attribution confidence
Very high

Why very high: the operator confirmed Union Alpha as Pareto 26.9 on 17 September 2026, 18 hours after the YFarmX post named it; the blend carries the Llama 3, Qwen and GLM tokenizers the measurements had read.

Evidence
  • ✓ near-exact agreement (46/47) with meta-llama/llama-3.2-3b-instruct
  • ✓ confirmed as Pareto 26.9, built by Circuit and Chisel by The Unbiased Co, 17 September 2026
Compare with Meta: Llama 3.2 3B InstructFull identity record

Ox Alpha

stealth/ox-alpha · Z.aiConfirmed by its maker
Confirmed family
Z.ai GLM
Confirmed as
GLM-5.3-Flash
Confirmed by Z.ai
26 August 2026
Entered catalogue
20 August 2026
Last tested
21 August 2026
Family attribution confidence
Very high

Why very high: Z.ai confirmed Ox Alpha as GLM-5.3-Flash on 26 August 2026, five days after the measurement matched it to GLM-5.3 on all fifty strings.

Evidence
  • ✓ exact tokenizer signature shared with 8 models
  • ✓ API-surface contract identical to a group member
  • ✓ confirmed as GLM-5.3-Flash by Z.ai, 26 August 2026
Compare with Z.ai: GLM 5.3 PrimeFull identity record

Space Bunny Alpha

stealth/space-bunny-alpha · Undisclosed (stealth)Family attributed by measurement
Likely family (measured)
MiniMax
Entered catalogue
23 September 2026
Last tested
23 September 2026
Family attribution confidence
High

Why high: an exact fingerprint shared with 8 models.

Evidence
  • ✓ exact tokenizer signature shared with 8 models
  • ○ awaiting the maker's or the platform's own word
Compare with MiniMax: MiniMax M3Full identity record

No family on the label

140 models whose catalogue entry names no tokenizer family. Here is what measuring each one says.
Placed
103
No match yet
15
Unmeasurable
6
Waiting to measure
16

Show:

ModelWhat the measurement saysHow sure we are

Builds on a known tokenizer

StepFun: Step 3.7 Flashstepfun/step-3.7-flashbuilt on the DeepSeek tokenizerHighStepFun: Step 3.5 Flashstepfun/step-3.5-flashbuilt on the DeepSeek tokenizerHighStepFun: Step 5 Previewstepfun/step-5-previewbuilt on the DeepSeek tokenizerHighGoogle: Gemma 3n 4Bgoogle/gemma-3n-e4b-itbuilt on the Gemini tokenizerHighThinking Machines: Inkling Smallthinkingmachines/inkling-smallbuilt on the GPT tokenizerVery highThinking Machines: Inklingthinkingmachines/inklingbuilt on the GPT tokenizerVery highInference.net: Schematron V2 Turboinference-net/schematron-v2-turbobuilt on the IBM tokenizerModerateIBM: Granite 4.2 8Bibm-granite/granite-4.2-8bbuilt on the Inference Net tokenizerModeratePareto 26.10 Previewunbiased/pareto-26.10-previewbuilt on the Llama3 tokenizerModerateInference.net: Schematron V2 Smallinference-net/schematron-v2-smallbuilt on the Llama3 tokenizerHighNous: Hermes 4 405Bnousresearch/hermes-4-405bbuilt on the Llama3 tokenizerVery highPerplexity: Sonar Properplexity/sonar-probuilt on the Llama3 tokenizerHighPerplexity: Sonarperplexity/sonarbuilt on the Llama3 tokenizerHighMeta: Muse Glimmer 30Bmeta/muse-glimmer-30bbuilt on the Llama4 tokenizerHighNVIDIA: Nemotron 3.5 Lightningnvidia/nemotron-3.5-lightningbuilt on the Mistral tokenizerHighNVIDIA: Nemotron 3 Ultranvidia/nemotron-3-ultra-550b-a55bbuilt on the Mistral tokenizerHighNVIDIA: Nemotron 3 Nano Omni (free)nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:freebuilt on the Mistral tokenizerHighNVIDIA: Nemotron 3 Supernvidia/nemotron-3-super-120b-a12bbuilt on the Mistral tokenizerHighNVIDIA: Nemotron 3 Nano 30B A3Bnvidia/nemotron-3-nano-30b-a3bbuilt on the Mistral tokenizerHighTheDrummer: Cydonia 24B V4.1thedrummer/cydonia-24b-v4.1built on the Mistral tokenizerModerateVenice: Uncensoredcognitivecomputations/dolphin-mistral-24b-venice-editionbuilt on the Mistral tokenizerHighTheDrummer: Skyfall 36B V2thedrummer/skyfall-36b-v2built on the Mistral tokenizerHighNVIDIA: Nemotron Nano 12B 2 VL (free)nvidia/nemotron-nano-12b-v2-vl:freebuilt on the Mistral tokenizerLowNVIDIA: Nemotron Nano 9B V2 (free)nvidia/nemotron-nano-9b-v2:freebuilt on the Mistral tokenizerHighFireworks: Ember-1fireworks/ember-1built on the Moonshot AI tokenizerHighIBM: Granite 4.0 Microibm-granite/granite-4.0-h-microbuilt on the OpenAI cl100k tokenizerHighMicrosoft: Phi 4microsoft/phi-4built on the OpenAI cl100k tokenizerHighIBM: Granite 4.1 8Bibm-granite/granite-4.1-8bbuilt on the OpenAI cl100k tokenizerHighApodex: Apodex 1.1 Mini (free)apodex/apodex-1.1-mini:freebuilt on the Qwen tokenizerHighPerceptron: Perceptron Mk1.5perceptron/perceptron-mk1.5built on the Qwen tokenizerHighXiaomi: MiMo-V2.6-Pro-UltraSpeedxiaomi/mimo-v2.6-pro-ultraspeedbuilt on the Qwen tokenizerHighXiaomi: MiMo-V2.6-Flashxiaomi/mimo-v2.6-flashbuilt on the Qwen tokenizerHighXiaomi: MiMo-V2.6-Proxiaomi/mimo-v2.6-probuilt on the Qwen tokenizerHighDots Studio: Dots3-Note Preview (free)dots-studio/dots-3-note-preview:freebuilt on the Qwen tokenizerHighPerceptron: Perceptron Mk1perceptron/perceptron-mk1built on the Qwen tokenizerHighXiaomi: MiMo-V2.5-Proxiaomi/mimo-v2.5-probuilt on the Qwen tokenizerHighXiaomi: MiMo-V2.5xiaomi/mimo-v2.5built on the Qwen tokenizerHighWriter: Palmyra X5writer/palmyra-x5built on the Qwen tokenizerHighRelace: Relace Searchrelace/relace-searchbuilt on the Qwen tokenizerHighByteDance: UI-TARS 7B bytedance/ui-tars-1.5-7bbuilt on the Qwen tokenizerHighMorph: Morph V3 Largemorph/morph-v3-largebuilt on the Qwen tokenizerHighMorph: Morph V3 Fastmorph/morph-v3-fastbuilt on the Qwen tokenizerHighKwaipilot: KAT-Coder-Pro V2kwaipilot/kat-coder-pro-v2built on the Qwen tokenizerHighKwaipilot: KAT-Coder-Air V2.5kwaipilot/kat-coder-air-v2.5built on the Qwen tokenizerHighAionLabs: Aion-3.0aion-labs/aion-3.0built on the Z.ai tokenizerModerate

Uses its own tokenizer

The fingerprint is shared only inside the maker's own model line: an in-house vocabulary rather than a borrowed one.

Z.ai: GLM 5.3 Primez-ai/glm-5.3-primeits own Z.ai tokenizer groupPlacedAionLabs: Aion 3.5 Miniaion-labs/aion-3.5-miniits own Aion Labs tokenizer groupPlacedAionLabs: Aion 3.5aion-labs/aion-3.5its own Aion Labs tokenizer groupPlacedUpstage: Solar Mini 4upstage/solar-mini4its own Upstage tokenizer groupPlacedZ.ai: GLM 5.3 FlashXz-ai/glm-5.3-flashxits own Z.ai tokenizer groupPlacedinclusionAI: Ling 3.0 Flash VLinclusionai/ling-3.0-flash-vlits own InclusionAI tokenizer groupPlacedinclusionAI: Ling 3.0 Flash Sante (free)inclusionai/ling-3.0-flash-sante:freeits own InclusionAI tokenizer groupPlacedinclusionAI: Ling 3.0 Flash Fininclusionai/ling-3.0-flash-finits own InclusionAI tokenizer groupPlacedZ.ai: GLM 5.3 Flashz-ai/glm-5.3-flashits own Z.ai tokenizer groupPlacedTencent: Hy-MT2-1.8Btencent/hy-mt2-1.8bits own Tencent tokenizer groupPlacedTencent: Hy-MT2-30B-A3Btencent/hy-mt2-30b-a3bits own Tencent tokenizer groupPlacedTencent: Hy-MT2-7Btencent/hy-mt2-7bits own Tencent tokenizer groupPlacedZ.ai: GLM 5.3z-ai/glm-5.3its own Z.ai tokenizer groupPlacedByteDance Seed: Seed 2.1 Turbobytedance-seed/seed-2-1-turboits own ByteDance Seed tokenizer groupPlacedByteDance Seed: Seed-2.0-Codebytedance-seed/seed-2.0-codeits own ByteDance Seed tokenizer groupPlacedUpstage: Solar Pro 4upstage/solar-pro4its own Upstage tokenizer groupPlacedinclusionAI: Ling 3.0 Flashinclusionai/ling-3.0-flashits own InclusionAI tokenizer groupPlacedMoonshotAI: Kimi K3moonshotai/kimi-k3its own Moonshot AI tokenizer groupPlacedAionLabs: Aion-3.0-Miniaion-labs/aion-3.0-miniits own Aion Labs tokenizer groupPlacedTencent: Hy3tencent/hy3its own Tencent tokenizer groupPlacedZ.ai: GLM 5.2z-ai/glm-5.2its own Z.ai tokenizer groupPlacedMoonshotAI: Kimi K2.7 Codemoonshotai/kimi-k2.7-codeits own Moonshot AI tokenizer groupPlacedNVIDIA: Nemotron 3.5 Content Safetynvidia/nemotron-3.5-content-safetyits own NVIDIA tokenizer groupPlacedMiniMax: MiniMax M3minimax/minimax-m3its own MiniMax tokenizer groupPlacedTencent: Hy3 previewtencent/hy3-previewits own Tencent tokenizer groupPlacedMoonshotAI: Kimi K2.6moonshotai/kimi-k2.6its own Moonshot AI tokenizer groupPlacedZ.ai: GLM 5.1z-ai/glm-5.1its own Z.ai tokenizer groupPlacedZ.ai: GLM 5V Turboz-ai/glm-5v-turboits own Z.ai tokenizer groupPlacedMiniMax: MiniMax M2.7minimax/minimax-m2.7its own MiniMax tokenizer groupPlacedZ.ai: GLM 5 Turboz-ai/glm-5-turboits own Z.ai tokenizer groupPlacedByteDance Seed: Seed-2.0-Litebytedance-seed/seed-2.0-liteits own ByteDance Seed tokenizer groupPlacedByteDance Seed: Seed-2.0-Minibytedance-seed/seed-2.0-miniits own ByteDance Seed tokenizer groupPlacedAionLabs: Aion-2.0aion-labs/aion-2.0its own Aion Labs tokenizer groupPlacedMiniMax: MiniMax M2.5minimax/minimax-m2.5its own MiniMax tokenizer groupPlacedZ.ai: GLM 5z-ai/glm-5its own Z.ai tokenizer groupPlacedMoonshotAI: Kimi K2.5moonshotai/kimi-k2.5its own Moonshot AI tokenizer groupPlacedUpstage: Solar Pro 3upstage/solar-pro-3its own Upstage tokenizer groupPlacedMiniMax: MiniMax M2-herminimax/minimax-m2-herits own MiniMax tokenizer groupPlacedZ.ai: GLM 4.7 Flashz-ai/glm-4.7-flashits own Z.ai tokenizer groupPlacedByteDance Seed: Seed 1.6 Flashbytedance-seed/seed-1.6-flashits own ByteDance Seed tokenizer groupPlacedByteDance Seed: Seed 1.6bytedance-seed/seed-1.6its own ByteDance Seed tokenizer groupPlacedMiniMax: MiniMax M2.1minimax/minimax-m2.1its own MiniMax tokenizer groupPlacedZ.ai: GLM 4.7z-ai/glm-4.7its own Z.ai tokenizer groupPlacedZ.ai: GLM 4.6Vz-ai/glm-4.6vits own Z.ai tokenizer groupPlacedMoonshotAI: Kimi K2 Thinkingmoonshotai/kimi-k2-thinkingits own Moonshot AI tokenizer groupPlacedMiniMax: MiniMax M2minimax/minimax-m2its own MiniMax tokenizer groupPlacedZ.ai: GLM 4.6z-ai/glm-4.6its own Z.ai tokenizer groupPlacedMoonshotAI: Kimi K2 0905moonshotai/kimi-k2-0905its own Moonshot AI tokenizer groupPlacedZ.ai: GLM 4.5Vz-ai/glm-4.5vits own Z.ai tokenizer groupPlacedZ.ai: GLM 4.5z-ai/glm-4.5its own Z.ai tokenizer groupPlacedZ.ai: GLM 4.5 Airz-ai/glm-4.5-airits own Z.ai tokenizer groupPlacedTencent: Hunyuan A13B Instructtencent/hunyuan-a13b-instructits own Tencent tokenizer groupPlacedMiniMax: MiniMax M1minimax/minimax-m1its own MiniMax tokenizer groupPlacedCohere: Command Acohere/command-aits own Cohere tokenizer groupPlacedMiniMax: MiniMax-01minimax/minimax-01its own MiniMax tokenizer groupPlacedinclusionAI: Ring-2.6-1Tinclusionai/ring-2.6-1tits own InclusionAI tokenizer groupPlacedinclusionAI: Ling-2.6-1Tinclusionai/ling-2.6-1tits own InclusionAI tokenizer groupPlacedinclusionAI: Ling-2.6-flashinclusionai/ling-2.6-flashits own InclusionAI tokenizer groupPlaced

Measured, no match yet

A clean fingerprint that no other measured model shares so far. Where something comes close, the nearest model is shown.

Paretounbiased/paretonothing close among the models measured so farNo match yetTencent: Hy4 previewtencent/hy4-previewnothing close among the models measured so farNo match yetLiquidAI: LFM2.5-2.6B (free)liquid/lfm-2.5-2.6b:freenothing close among the models measured so farNo match yetPoolside: Laguna S 2.1poolside/laguna-s-2.1nothing close among the models measured so farNo match yetMeituan: LongCat 2.0meituan/longcat-2.0nothing close among the models measured so farNo match yetPoolside: Laguna XS 2.1poolside/laguna-xs-2.1nothing close among the models measured so farNo match yetArcee AI: Trinity Large Thinkingarcee-ai/trinity-large-thinkingnothing close among the models measured so farNo match yetGoogle: Lyria 3 Clip Previewgoogle/lyria-3-clip-previewnothing close among the models measured so farNo match yetReka Edgerekaai/reka-edgenothing close among the models measured so farNo match yetMoonshotAI: Kimi K2 0711moonshotai/kimi-k2nothing close among the models measured so farNo match yetBaidu: ERNIE 4.5 VL 424B A47B baidu/ernie-4.5-vl-424b-a47bnothing close among the models measured so farNo match yetMeta: Llama Guard 4 12Bmeta-llama/llama-guard-4-12bnothing close among the models measured so farNo match yetReka Flash 3rekaai/reka-flash-3nothing close among the models measured so farNo match yetAionLabs: Aion-RP 1.0 (8B)aion-labs/aion-rp-llama-3.1-8bnothing close among the models measured so farNo match yetTypeSafe: Jev 1.13typesafe/jev-1.13nothing close among the models measured so farNo match yet

Could not be measured reliably

Their providers report token counts too disturbed (by caching or shifting overheads) to trust. Publishing a corrupted fingerprint would be worse than none.

inclusionAI: Ling 3.1 Flashinclusionai/ling-3.1-flashtoken accounting unreliable at this endpointUnmeasurableSakana: Fugu Maxsakana/fugu-maxtoken accounting unreliable at this endpointUnmeasurableInception: Mercury 2.5inception/mercury-2.5token accounting unreliable at this endpointUnmeasurableGoogle: Lyria 3 Pro Previewgoogle/lyria-3-pro-previewtoken accounting unreliable at this endpointUnmeasurableInception: Mercury 2inception/mercury-2token accounting unreliable at this endpointUnmeasurableInception: Mercury 2.5 Previewinception/mercury-2.5-previewtoken accounting unreliable at this endpointUnmeasurable

Not measured yet

Queued for the next measurement run; results land here as each one completes.

Sakana: Fugu Ultra v2sakana/fugu-ultra-v2in the measurement queueQueuedMeta: Muse Spark 1.3 Contributormeta/muse-spark-1.3-contributorin the measurement queueQueuedMeta: Muse Spark 1.3meta/muse-spark-1.3in the measurement queueQueuedMeta: Muse Spark 1.2 Contributormeta/muse-spark-1.2-contributorin the measurement queueQueuedSakana: Sakana Namazusakana/sakana-namazuin the measurement queueQueuedMeta: Muse Spark 1.2meta/muse-spark-1.2in the measurement queueQueuedMeta: Muse Spark 1.1meta/muse-spark-1.1in the measurement queueQueuedKwaipilot: KAT-Coder-Pro V2.5kwaipilot/kat-coder-pro-v2.5in the measurement queueQueuedSakana: Fugu Ultrasakana/fugu-ultrain the measurement queueQueuedPerplexity: Sonar Pro Searchperplexity/sonar-pro-searchin the measurement queueQueuedRelace: Relace Apply 3relace/relace-apply-3in the measurement queueQueuedPerplexity: Sonar Reasoning Properplexity/sonar-reasoning-proin the measurement queueQueuedPerplexity: Sonar Deep Researchperplexity/sonar-deep-researchin the measurement queueQueuedAllenAI: Olmo 3 32B Thinkallenai/olmo-3-32b-thinkin the measurement queueQueuedDeep Cogito: Cogito v2.1 671Bdeepcogito/cogito-v2.1-671bin the measurement queueQueuedArcee AI: Virtuoso Largearcee-ai/virtuoso-largein the measurement queueQueued

Why measure models everyone knows?

Because the known models are the reference library. A stealth model can only be placed by comparing its fingerprint against models whose families are established, so every known model measured makes the next unknown easier to identify. That is exactly how Ox Alpha was placed within minutes of measurement. Method and limits: the methodology page.