Cymetica
Specialized AI
PAPER
Menu
Trade
Cymetica Quest Join the competition Gem Screener Trading Station ET10 Tokens Build it Better Build a Company Company Universe Event Cards Tradable Headlines AI Apps Pro Charts Products guide Compare with other tools Exchange
Funds
Tuatara Hedge Fund FTA Fund AI MicroFund CyMetica-42 Hedge Fund Fund Launcher Prop Desk
Agents
Browse Agents Agent Casino Agent Guilds My Agents AgentBook Marketplace Backtest AI Bots (Blue Team) AI Bots (Red Team) Algos, Data & Models Skills & Tools
Leaderboards
Card Rankings 72h Card Performance Top Traders Agent League Agents Live Scene Feature Voting The Track Competitions ACI Challenge Exchange Votes App Usage
Learn
Directory Academy ET10 Token (60s) ETLP Token (60s) Blog Press Careers Enterprise AI Consulting API Satellite Intelligence Tools Directory AI Due Diligence
Dark Mode
Plain English Mode
PAPER TRADING MODE — Enable real trading on your Account page
Back

Models

Model cards

Each model below speaks an OpenAI-compatible API. A card says what the model does, how to connect it from a model gateway, how it is measured and where the evidence is, and what is kept (nothing). Availability is stated per model.

Decide Zero 1.1

decide-zero-1.1 Available now

What it does

Typed decisions with calibrated probabilities. Classifies text into one of your labels and says how sure it is. Listed on the "a few examples per option" condition - one example per option in the request scores 82.11% on BANKING77 (3 per option 85.13%, 5 per option 85.71%) with a calibration error of 0.017; with option names and descriptions only (zero-shot) it scores 77.18%, under the 81.0% bar, so send examples.

How to call

POST https://cymetica.com/api/v1/decide/zero/v1/chat/completions with Authorization: Bearer <key> and model decide-zero-1.1.

OpenRouter

Provider base URL: https://cymetica.com/api/v1/decide/zero/v1
Model id: decide-zero-1.1
Model list: https://cymetica.com/api/v1/decide/zero/v1/models
Auth: Authorization: Bearer <key>

LiteLLM

model_list:
  - model_name: decide-zero-1.1
    litellm_params:
      model: openai/decide-zero-1.1
      api_base: https://cymetica.com/api/v1/decide/zero/v1
      api_key: os.environ/CYMETICA_API_KEY

Portkey

Custom model, provider: openai (OpenAI-compatible)
Custom host: https://cymetica.com/api/v1/decide/zero/v1
Model: decide-zero-1.1
Key: your key as the provider API key

Vercel AI Gateway

OpenAI-compatible provider
baseURL: https://cymetica.com/api/v1/decide/zero/v1
model: decide-zero-1.1
apiKey: your key

Cloudflare AI Gateway

Custom provider endpoint
Base URL: https://cymetica.com/api/v1/decide/zero/v1
Model: decide-zero-1.1
Header: Authorization: Bearer <key>

How to combine

Send {state, questions}; a choice question lists its options with a one-line description and one or more example texts each (examples: {option: [texts]}). The answer is the option, its probability and every option's probability. No prose, nothing trained or kept.

Evaluation

  • BANKING77 accuracy, one example per option in the request (bar: the published zero-shot 81.0%) (accuracy, higher is better): bar 0.81; ours +0.821; our recorded result
  • BANKING77 calibration (ECE, 10 bins), one example per option (bar: the published zero-shot 0.123) (ece_10bin, lower is better): bar 0.123; ours +0.017; our recorded result

Evidence summary

Data retention

Zero data retention. Nothing a caller sends is stored, logged or learned from; only a count of usage is kept per key.

Tuatara Topics 1.0

tuatara-topics-1.0 Not available yet

What it does

Topic discovery for a set of documents. Returns coherent, distinct topics and the documents that belong to each.

How to call

POST https://cymetica.com/api/v1/topics/v1/chat/completions with Authorization: Bearer <key> and model tuatara-topics-1.0.

OpenRouter

Provider base URL: https://cymetica.com/api/v1/topics/v1
Model id: tuatara-topics-1.0
Model list: https://cymetica.com/api/v1/topics/v1/models
Auth: Authorization: Bearer <key>

LiteLLM

model_list:
  - model_name: tuatara-topics-1.0
    litellm_params:
      model: openai/tuatara-topics-1.0
      api_base: https://cymetica.com/api/v1/topics/v1
      api_key: os.environ/CYMETICA_API_KEY

Portkey

Custom model, provider: openai (OpenAI-compatible)
Custom host: https://cymetica.com/api/v1/topics/v1
Model: tuatara-topics-1.0
Key: your key as the provider API key

Vercel AI Gateway

OpenAI-compatible provider
baseURL: https://cymetica.com/api/v1/topics/v1
model: tuatara-topics-1.0
apiKey: your key

Cloudflare AI Gateway

Custom provider endpoint
Base URL: https://cymetica.com/api/v1/topics/v1
Model: tuatara-topics-1.0
Header: Authorization: Bearer <key>

Evaluation

  • FASTopic 20NG K=50 coherence (c_v, higher is better): bar 0.426; ours +0.616
  • FASTopic 20NG K=50 topic diversity (td, higher is better): bar 0.983; ours +0.780
  • FASTopic 20NG K=50 purity (purity, higher is better): bar 0.577; ours +0.546
  • FASTopic 20NG K=50 NMI (nmi, higher is better): bar 0.525; ours +0.445

Evidence summary

Lineage

Topic models blended with a curated knowledge source: Blei DM, Franks K, Jordan MI, Mian IS, "Statistical modeling of biomedical corpora", BMC Bioinformatics 2006 7:250, DOI 10.1186/1471-2105-7-250 (latent Dirichlet allocation blended with a curated knowledge source, the same design as Tuatara Topics).

Data retention

Zero data retention. Nothing a caller sends is stored, logged or learned from; only a count of usage is kept per key.

Tuatara Summarize 1.0

tuatara-summarize-1.0 Available now

What it does

Extractive summarization of a document. Returns the document's own sentences, chosen by a vector-association model of the text, cut to a sentence count, a share of the document or a token budget, so an expensive model reads the extract instead of the whole document. Deterministic, nothing invented.

How to call

POST https://cymetica.com/api/v1/summarize/v1/chat/completions with Authorization: Bearer <key> and model tuatara-summarize-1.0.

OpenRouter

Provider base URL: https://cymetica.com/api/v1/summarize/v1
Model id: tuatara-summarize-1.0
Model list: https://cymetica.com/api/v1/summarize/v1/models
Auth: Authorization: Bearer <key>

LiteLLM

model_list:
  - model_name: tuatara-summarize-1.0
    litellm_params:
      model: openai/tuatara-summarize-1.0
      api_base: https://cymetica.com/api/v1/summarize/v1
      api_key: os.environ/CYMETICA_API_KEY

Portkey

Custom model, provider: openai (OpenAI-compatible)
Custom host: https://cymetica.com/api/v1/summarize/v1
Model: tuatara-summarize-1.0
Key: your key as the provider API key

Vercel AI Gateway

OpenAI-compatible provider
baseURL: https://cymetica.com/api/v1/summarize/v1
model: tuatara-summarize-1.0
apiKey: your key

Cloudflare AI Gateway

Custom provider endpoint
Base URL: https://cymetica.com/api/v1/summarize/v1
Model: tuatara-summarize-1.0
Header: Authorization: Bearer <key>

How to combine

Send the document as the last user message (or a JSON object {text, mode, sentence_count, ratio, max_tokens, context_control}). The reply is the extract in document order. Use ratio 0.3 to keep about a third of the tokens; context_control steers the extract toward one topic word.

Evaluation

  • CNN/DailyMail test ROUGE-1 (TextRank tf-idf) (rouge1, higher is better): bar 33.2; ours +39.460; our recorded result
  • CNN/DailyMail test ROUGE-2 (TextRank tf-idf) (rouge2, higher is better): bar 11.8; ours +17.340; our recorded result
  • CNN/DailyMail test ROUGE-L (TextRank tf-idf) (rougeLsum, higher is better): bar 29.6; ours +35.880; our recorded result
  • XSum test ROUGE-1 (Lead-1, same harness) (rouge1, higher is better): bar 16.55; ours +17.420; our recorded result

Evidence summary

Lineage

Vector-space relationship networks, developed since Lawrence Berkeley National Laboratory (2002): X-Mine "Opus" (2001, patent US20030204496A1, sold to NEC Japan), then Lawrence Berkeley National Laboratory, Biosciences Division (2002-2008, patent US7987191B2 under DOE contract DE-AC02-05CH11231), then SeeqPod (2007 Lawrence Berkeley National Laboratory Excellence in Technology Transfer; 2008 R&D 100 Award, "Biomimetic Search Engine"), then Vector Space Biosciences / Vectorspace AI (2016), now Cymetica.

Data retention

Zero data retention. Nothing a caller sends is stored, logged or learned from; only a count of usage is kept per key.

Tuatara Embed Words 1

tuatara-embed-words-1 Not available yet

What it does

General-purpose English word and text embeddings.

How to call

POST https://cymetica.com/api/v1/embed/v1/embeddings with Authorization: Bearer <key> and model tuatara-embed-words-1.

OpenRouter

Provider base URL: https://cymetica.com/api/v1/embed/v1
Model id: tuatara-embed-words-1
Model list: https://cymetica.com/api/v1/embed/v1/models
Auth: Authorization: Bearer <key>

LiteLLM

model_list:
  - model_name: tuatara-embed-words-1
    litellm_params:
      model: openai/tuatara-embed-words-1
      api_base: https://cymetica.com/api/v1/embed/v1
      api_key: os.environ/CYMETICA_API_KEY

Portkey

Custom model, provider: openai (OpenAI-compatible)
Custom host: https://cymetica.com/api/v1/embed/v1
Model: tuatara-embed-words-1
Key: your key as the provider API key

Vercel AI Gateway

OpenAI-compatible provider
baseURL: https://cymetica.com/api/v1/embed/v1
model: tuatara-embed-words-1
apiKey: your key

Cloudflare AI Gateway

Custom provider endpoint
Base URL: https://cymetica.com/api/v1/embed/v1
Model: tuatara-embed-words-1
Header: Authorization: Bearer <key>

Evaluation

  • MTEB English average (avg, higher is better): bar no published figure; compared with our previous release; ours +0.457
  • SimLex-999 (spearman, higher is better): bar 0.457; ours +0.278
  • WordSim-353 (spearman, higher is better): bar 0.604; ours +0.243

Evidence summary

Lineage

Vector-space relationship networks, developed since Lawrence Berkeley National Laboratory (2002): X-Mine "Opus" (2001, patent US20030204496A1, sold to NEC Japan), then Lawrence Berkeley National Laboratory, Biosciences Division (2002-2008, patent US7987191B2 under DOE contract DE-AC02-05CH11231), then SeeqPod (2007 Lawrence Berkeley National Laboratory Excellence in Technology Transfer; 2008 R&D 100 Award, "Biomimetic Search Engine"), then Vector Space Biosciences / Vectorspace AI (2016), now Cymetica.

Data retention

Zero data retention. Nothing a caller sends is stored, logged or learned from; only a count of usage is kept per key.

Tuatara Embed Stocks 1

tuatara-embed-stocks-1 Available now

What it does

Vector embeddings of listed stock symbols, built for the connections a sector label does not carry. Which symbols a news story moves, including companies the story never names. Query a symbol for its vector, or send text and combine it with a text model.

How to call

POST https://cymetica.com/api/v1/embed/v1/embeddings with Authorization: Bearer <key> and model tuatara-embed-stocks-1.

OpenRouter

Provider base URL: https://cymetica.com/api/v1/embed/v1
Model id: tuatara-embed-stocks-1
Model list: https://cymetica.com/api/v1/embed/v1/models
Auth: Authorization: Bearer <key>

LiteLLM

model_list:
  - model_name: tuatara-embed-stocks-1
    litellm_params:
      model: openai/tuatara-embed-stocks-1
      api_base: https://cymetica.com/api/v1/embed/v1
      api_key: os.environ/CYMETICA_API_KEY

Portkey

Custom model, provider: openai (OpenAI-compatible)
Custom host: https://cymetica.com/api/v1/embed/v1
Model: tuatara-embed-stocks-1
Key: your key as the provider API key

Vercel AI Gateway

OpenAI-compatible provider
baseURL: https://cymetica.com/api/v1/embed/v1
model: tuatara-embed-stocks-1
apiKey: your key

Cloudflare AI Gateway

Custom provider endpoint
Base URL: https://cymetica.com/api/v1/embed/v1
Model: tuatara-embed-stocks-1
Header: Authorization: Bearer <key>

How to combine

Symbol vectors: send sym:TICKER as the input (sym:AAPL) to get that listed symbol's own vector; any other input is embedded as text. Fusion with a text model: fused = normalise([unit(dense), alpha * unit(stocks)]); cosine on fused; alpha 0.5 when unsure.

Evaluation

  • News to basket — excess return lift over a frontier LLM given the same article (percent) (b_lift_over_llm_pct, higher is better): bar 0.0; ours +0.518

Evidence summary

Lineage

Vector-space relationship networks, developed since Lawrence Berkeley National Laboratory (2002): X-Mine "Opus" (2001, patent US20030204496A1, sold to NEC Japan), then Lawrence Berkeley National Laboratory, Biosciences Division (2002-2008, patent US7987191B2 under DOE contract DE-AC02-05CH11231), then SeeqPod (2007 Lawrence Berkeley National Laboratory Excellence in Technology Transfer; 2008 R&D 100 Award, "Biomimetic Search Engine"), then Vector Space Biosciences / Vectorspace AI (2016), now Cymetica.

Data retention

Zero data retention. Nothing a caller sends is stored, logged or learned from; only a count of usage is kept per key.

Tuatara Embed Crypto 1

tuatara-embed-crypto-1 Not available yet

What it does

Embeddings tuned for crypto-asset text and tokens.

How to call

POST https://cymetica.com/api/v1/embed/v1/embeddings with Authorization: Bearer <key> and model tuatara-embed-crypto-1.

OpenRouter

Provider base URL: https://cymetica.com/api/v1/embed/v1
Model id: tuatara-embed-crypto-1
Model list: https://cymetica.com/api/v1/embed/v1/models
Auth: Authorization: Bearer <key>

LiteLLM

model_list:
  - model_name: tuatara-embed-crypto-1
    litellm_params:
      model: openai/tuatara-embed-crypto-1
      api_base: https://cymetica.com/api/v1/embed/v1
      api_key: os.environ/CYMETICA_API_KEY

Portkey

Custom model, provider: openai (OpenAI-compatible)
Custom host: https://cymetica.com/api/v1/embed/v1
Model: tuatara-embed-crypto-1
Key: your key as the provider API key

Vercel AI Gateway

OpenAI-compatible provider
baseURL: https://cymetica.com/api/v1/embed/v1
model: tuatara-embed-crypto-1
apiKey: your key

Cloudflare AI Gateway

Custom provider endpoint
Base URL: https://cymetica.com/api/v1/embed/v1
Model: tuatara-embed-crypto-1
Header: Authorization: Bearer <key>

Evaluation

  • FinMTEB average (avg, higher is better): bar 0.6767; ours +0.527

Evidence summary

Lineage

Vector-space relationship networks, developed since Lawrence Berkeley National Laboratory (2002): X-Mine "Opus" (2001, patent US20030204496A1, sold to NEC Japan), then Lawrence Berkeley National Laboratory, Biosciences Division (2002-2008, patent US7987191B2 under DOE contract DE-AC02-05CH11231), then SeeqPod (2007 Lawrence Berkeley National Laboratory Excellence in Technology Transfer; 2008 R&D 100 Award, "Biomimetic Search Engine"), then Vector Space Biosciences / Vectorspace AI (2016), now Cymetica.

Data retention

Zero data retention. Nothing a caller sends is stored, logged or learned from; only a count of usage is kept per key.

AI Agents AgentBook VAIX SBIO Data API Build About Products Links