Models Deca 2.5 Mini Deca 2.5 Pro Deca 2.5 Ultra
Usage Guide Get Your API Key Fire Up a Request Parameters Cheat Sheet Counting (and Saving) Tokens Turbo & Dwarf Modes Error Codes That Matter

Models

Below are the list of models from Deca:

Latest models (2.5)

Deca 2.5 Pro

Intelligence:

Cost:

  • Normal Price:

    $1.25 / MTok in, $4 / MTok out

  • Turbo Price:

    $2 / MTok in, $5.50 / MTok out

Reasoning:

  • Sometimes
    Deca dynamically choses when to think based on the prompt.

Context Length:

  • 128000 Context + 32000 Attention Context
  • 160000 tokens total. Deca's architecture is designed to allocate more attention heads to the 32000 attention context tokens than the 128000 normal context tokens. Deca intelligently chooses which tokens to attend to based on the prompt. This usually results in a better performance. Disable this by using `attention_density_adjustment=0` in the request body

Deca 2.5 pro delivers a great balance between performance and cost.

Snapshots:
  • deca-2.5-pro:flex

    ? / MTok in, ? / MTok out

    Flex tier allows you to spend less by only paying for the GPU resources you use. It is generally cheaper than the normal tier. Requests might be queued for processing. Prompts will be used to improve Deca's AI models
  • deca-2.5-pro:free

    $0 / MTok in, $0 / MTok out

    Due to extremely high demand, this snapshot is currently unavailable. We are working to improve rate limits and usage for all our customers.
  • deca-2.5-pro
  • deca-2.5-pro-turbo
  • deca-2.5-pro-beta-06032025
Maximum price with all dwarves activated $0.8 / MTok in, $2.2 / MTok out

Deca 2.5 Ultra Preview

Intelligence:

Cost:

  • Normal Price:

    $25 / MTok in, $50 / MTok out

    Not availble yet.

  • Preview Price:

    $75 / MTok in, $150 / MTok out

Reasoning:

  • Always

Context Length:

  • 1000000 context

Deca 2.5 Ultra is our largest, and most intelligent model in the Deca 2.5 series. BETA

Deca 2.5 is currently in beta and is subject to change.
Snapshots:
  • deca-2.5-ultra
  • deca-2.5-ultra-07012025

Deca 2.5 mini

Intelligence:

Cost:

  • Normal Price:

    $0.35 / MTok in, $0.99 / MTok out

  • Turbo Price:

    $0.99 / MTok in, $0.99 / MTok out

Reasoning:

  • No

Context Length:

  • 128000 context

Deca 2.5 mini is our smallest model in the Deca 2.5 series.

Snapshots:
  • deca-2.5-mini:flex

    ? / MTok in, ? / MTok out

    Flex tier allows you to spend less by only paying for the GPU resources you use. It is generally cheaper than the normal tier. Requests might be queued for processing. Prompts will be used to improve Deca's AI models
  • deca-2.5-mini
  • deca-2.5-mini-turbo
  • deca-2.5-mini-beta-06032025

Using Deca Models

So you’ve picked your model (great choice, by the way) – now what? Below is the quick-start guide to using Deca models

1. Get Your API Key

Sign in to the and head to the API Keys page (Key icon in your bottom bar). Smash that green button, copy the key and paste it into the code below. Keep it safe.

Don't have an account? for free.

2. Fire Up a Request

curl https://api.deca.ai/v1/chat/completions \
  -H "Authorization: Bearer <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deca-2.5-ultra-max",
    "messages": [
      { "role": "user", "content": "Write acrostic about Deca, a powerful AI model" }
    ]
  }'

# Expected response:
# Dynamic intelligence, ever-learning with grace,
# Empowering innovation, leading the race.
# Connecting minds, shaping the future bright,
# Always adapting, a beacon of light.

Substitute deca-2.5-ultra-max with any model or snapshot from the table above.

3. Parameters Cheat-Sheet

ParamWhat It DoesDefault
temperatureSpices up randomness. 0 = robotic, 2 = poet on espresso.0.7
max_tokensHow long the model can talk before we shush it.Max (see models)
top_pTop_p controls text diversity by selecting from the smallest set of words whose probabilities add up to a threshold. 1.0
streamSet to true for token-by-token response.false

4. Counting (and Saving) Tokens

Prices are shown as $ / MTok – that’s per million tokens. For a ten-token “What are the best hotels in Italy?”, you’ll pay roughly the cost of finding a penny on the sidewalk. Use the cheaper :free snapshots for testing, then move up when you feel you're ready for production.

5. Turbo & Dwarf Modes

Add -turbo to any model name for lightning-fast responses (at a slightly higher price). Ultra models: use higher reasoning efforts to activate more dwarves. Use deca-2.5-ultra-max to activate all dwarves.

6. Error Codes That Matter