Google: Gemma 4 12B Starry Ocean

gemma-4-12b-it-starry-ocean

Gemma 4 12B Starry Ocean is an open-weight merge for storytelling, roleplay, and creative writing. It supports image understanding, tool calling, optional reasoning, and a 131,072-token context window.

Modalities

Text + Image → Text

Input / output price

$0.065 / $0.325 per 1M

Context

131K

Capabilities

RoleplayReasoningVisionTool callingOpen weights

Pricing

Billed per token actually used and deducted from your credits.

ItemUSD per 1M tokens
Input$0.065
Output$0.325
Cached input$0.0325

Calculator

Drag the sliders to estimate what this model will cost.

Per request
$0.00046
Per day
$0.2275
Per month (30 days)
$6.83

Gemma 4 12B Starry Ocean: input $0.065/M · output $0.325/M · cached input $0.0325/M

Request with these settings
curl https://api.geniusrelaygpt.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemma-4-12b-it-starry-ocean", "max_tokens": 800,
       "messages": [{"role": "user", "content": "…"}]}'

Code samples

OpenAI-compatible. Set the base URL to https://api.geniusrelaygpt.com/v1 and the model to gemma-4-12b-it-starry-ocean.

curl https://api.geniusrelaygpt.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-4-12b-it-starry-ocean",
    "stream": true,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Use it in a client: Cursor · Cline · SillyTavern · Codex

Parameters

Common OpenAI parameters can be passed as-is.

temperaturetop_pmax_tokensstreamstoppresence_penaltyfrequency_penaltyseed

FAQ

How is Gemma 4 12B Starry Ocean billed?
Input is $0.065 and output is $0.325 per million tokens; cached input is $0.0325. The charge for every request is listed in Usage Logs in the console.
How long is the context window?
131K tokens. In a long conversation every turn is billed for the full input, so keep the history under control.
Does it support streaming?
Yes. Add "stream": true to the request to receive server-sent events in the OpenAI format.