AI Desk · AI Fingerprinting

Identity watch

A model can change behind its name. The watch records what appeared, what measured differently, and what measurement placed, so a silent change stops being silent.

Price watch

What the labs charge, as listed on OpenRouter on each snapshot date, in dollars per million tokens: the median of each lab's priced text models, prompt price first and completion price second

Lab	21 Aug	3 Sept	16 Sept	5 Oct	8 Oct
Every priced text model320 models on 21 Aug, 349 on 8 Oct	$0.428$1.60	$0.428$1.70	$0.40$1.738	$0.40$1.775	$0.425$1.80
OpenAI65 priced models	$1.75$10.00	$1.75$10.00	$1.75$10.00	$2.00$10.00	$2.00$10.00
Anthropic16 priced models	$5.00$25.00	$5.00$25.00	$5.00$25.00	$5.00$25.00	$4.50$22.50
Google27 priced models	$0.375$2.50	$0.50$3.00	$0.50$2.75	$0.50$2.75	$0.50$3.00
DeepSeek14 priced models	$0.27$0.975	$0.295$0.945	$0.269$0.95	$0.27$1.084	$0.264$1.014
Alibaba (Qwen)53 priced models	$0.22$1.162	$0.21$1.125	$0.20$1.125	$0.20$1.10	$0.21$1.10
Z.ai16 priced models	$0.60$2.00	$0.60$2.06	$0.60$1.86	$0.515$2.06	$0.515$2.06
Moonshot AI7 priced models	$0.60$2.50	$0.60$2.50	$0.60$2.50	$0.60$2.50	$0.60$2.50
xAI7 priced models	$1.25$2.50	$1.25$2.50	$1.25$2.50	$1.25$2.50	$1.25$2.50
Mistral AI20 priced models	$0.20$0.60	$0.30$0.60	$0.30$0.60	$0.30$0.60	$0.326$0.75
Meta8 priced models	$0.10$0.31	$0.10$0.31	$0.10$0.31	$0.14$0.315	$0.10$0.31
MiniMax8 priced models	$0.30$1.20	$0.30$1.20	$0.30$1.20	$0.30$1.20	$0.30$1.20
Xiaomi5 priced models	under 3 models	under 3 models	under 3 models	$0.435$0.87	$0.435$0.87

A lab's median moves when it adds or withdraws models as well as when it reprices one, so the list below names the individual moves.

Biggest price moves, 5 Oct to 8 Oct

39 listed prices moved in those 3 days: 16 up and 23 down. The twelve largest, by percentage:

Qwen: Qwen3 30B A3B Instruct 2507qwen/qwen3-30b-a3b-instruct-2507Prompt price: $0.048/M became $0.10/M+107.7%DeepSeek: DeepSeek V4.1 Flashdeepseek/deepseek-v4.1-flashPrompt price: $0.30/M became $0.028/M-90.6%DeepSeek: DeepSeek V4 Flash 0423deepseek/deepseek-v4-flashPrompt price: $0.03/M became $0.007/M-75%Z.ai: GLM 5.3z-ai/glm-5.3Prompt price: $0.05/M became $0.085/M+70%Qwen: Qwen3.5-27Bqwen/qwen3.5-27bCompletion price: $1.56/M became $2.60/M+66.7%DeepSeek: DeepSeek V4 Pro 0813deepseek/deepseek-v4-pro-0813Prompt price: $0.40/M became $0.66/M · base rate; priced by time of day+65%DeepSeek: DeepSeek V4 Pro 0813deepseek/deepseek-v4-pro-0813Completion price: $5.00/M became $1.98/M · base rate; priced by time of day-60.4%Qwen: Qwen3 235B A22B Instruct 2507qwen/qwen3-235b-a22b-2507Completion price: $0.35/M became $0.55/M+57.1%Z.ai: GLM 5.2z-ai/glm-5.2Prompt price: $0.019/M became $0.03/M+56.2%Qwen: Qwen3 30B A3B Instruct 2507qwen/qwen3-30b-a3b-instruct-2507Completion price: $0.193/M became $0.30/M+55.4%Meta: Llama 3.3 70B Instructmeta-llama/llama-3.3-70b-instructPrompt price: $0.22/M became $0.10/M-54.5%MoonshotAI: Kimi K2.6moonshotai/kimi-k2.6Prompt price: $0.95/M became $0.438/M-53.9%

What changed

Every row is a measured difference between two dated snapshots: the value it was, and the value it became

5 October 2026

16 September 2026

3 September 2026

21 August 2026

Sourced from the frozen catalogue snapshots of 20 August 2026 and 21 August 2026 and 3 September 2026 and 16 September 2026 and 5 October 2026. As more snapshots freeze, this timeline grows one dated interval at a time.

New in the catalogue

Entries that appeared between the snapshots of 16 September 2026 and 8 October 2026. Catalogue appearance is not a release date: a model can join the catalogue long after it shipped.

Gone from the catalogue

59 entries the catalogue no longer lists. Their records and measurements stay published: a model disappearing is a change worth keeping.

Stealth watch

Undeclared models, measured while their identity stays hidden

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 Prime Full identity record

Space Bunny Alphastealth/space-bunny-alphaMiniMax family candidateStealth

New lineage placements

Models whose catalogue entry names no family, placed for the first time by measurement

Apodex: Apodex 1.1 Mini (free)apodex/apodex-1.1-mini:freeBuilt on the Qwen tokenizerPlaced Perceptron: Perceptron Mk1.5perceptron/perceptron-mk1.5Built on the Qwen tokenizerPlaced Fireworks: Ember-1fireworks/ember-1Built on the Moonshot AI tokenizerPlaced Z.ai: GLM 5.3 Primez-ai/glm-5.3-primeIts own Z.ai tokenizer groupPlaced AionLabs: Aion 3.5 Miniaion-labs/aion-3.5-miniIts own Aion Labs tokenizer groupPlaced AionLabs: Aion 3.5aion-labs/aion-3.5Its own Aion Labs tokenizer groupPlaced Upstage: Solar Mini 4upstage/solar-mini4Its own Upstage tokenizer groupPlaced Xiaomi: MiMo-V2.6-Pro-UltraSpeedxiaomi/mimo-v2.6-pro-ultraspeedBuilt on the Qwen tokenizerPlaced Xiaomi: MiMo-V2.6-Flashxiaomi/mimo-v2.6-flashBuilt on the Qwen tokenizerPlaced Xiaomi: MiMo-V2.6-Proxiaomi/mimo-v2.6-proBuilt on the Qwen tokenizerPlaced Z.ai: GLM 5.3 FlashXz-ai/glm-5.3-flashxIts own Z.ai tokenizer groupPlaced Inference.net: Schematron V2 Turboinference-net/schematron-v2-turboBuilt on the IBM tokenizerPlaced

How the watch works

Each sweep freezes a dated catalogue snapshot and appends new measurements; nothing is overwritten. A new entry is fingerprinted and compared against the whole corpus, a fingerprint change is re-tested before it is published, and a model that leaves the catalogue keeps its record. When identity alerts open, a watched model will be able to notify you; until then, this board and each model's history carry the changes.