NOUMENOpen app
NOUMEN

Google · Open weights

Gemma 4 12B Starry Ocean

Added Oct 15, 2026Open weights

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Use in the app

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.

  • Reasoning
  • Tools
  • Vision
Input
Text, Images
Output
Text
Reasoning effort
none · high
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.05
Output
$0.25
Cache read
$0.025

Reference prices as published by the model provider. API example

Specifications

Context window
131.1K
Max output
—
Average output (7 days)
233 tokens
Longer than 25% of 473 models

Side by side

Gemma 4 12B Starry Ocean compared

Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Gemma 4 12B Starry OceanGoogleThis modelDeepSeek V4 Flash 0731 (Thinking)DeepSeekRemove
ProviderGoogleDeepSeek
Discovery tierOpen weightsOpen weights
AddedOct 15, 2026Aug 1, 2026
Limits
Context window131.07K1.05MBest
Max output—131.07K
Average output233 tokens755 tokens
Pricing, USD per 1M tokens
Input$0.05$0.05
Output$0.25$0.16Best
Cache read$0.025$0.013Best
Performance
Throughput18.1 tok/s151.9 tok/sBest
Time to first token1.8 sBest7.0 s
Request success (24h)99.3%99.7%Best
Benchmark indexes
Intelligence Index—34.3
Coding Index—69.1
Capabilities
ReasoningYesYes
ToolsYesYes
VisionYesNo
Structured outputNoYes
Reasoning effortnone, highnone, low, high, max

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

  • Gemma 4 12B Starry Ocean
  • DeepSeek V4 Flash 0731 (Thinking)

Intelligence Index

  • Gemma 4 12B Starry Ocean—
  • DeepSeek V4 Flash 0731 (Thinking)34.3
3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 69.1

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 54.0%

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 23.0%

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 1,257 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 1,433 Elo

Document reasoning

  • GDP.pdf

    Professional reasoning over PDF documents, all-pass rate

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 11.0%

  • AA-LCRv1.1

    Reasoning across long documents

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 79.7%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 38.6%

  • GPQA Diamond

    Graduate-level science questions

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 90.8%

Coding

  • Terminal-Benchv4.0

    Practical software engineering in a terminal

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 12.1%

  • SciCode

    Scientific computing in Python

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 50.3%

Legacy evaluations

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Gemma 4 12B Starry Ocean: —
    • DeepSeek V4 Flash 0731 (Thinking): 79.7%

Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Throughput
18.1 tok/s
Faster than 9% of 256 models
Time to first token
1.8 s
Quicker than 78% of 256 models
Prompt cache hits
69.2%
About $0.0327 per 1M input tokens with caching
Request success (24h)
99.3%
Reliable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 88.8%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-4-12b-it-starry-ocean",
    "reasoning_effort": "none",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="gemma-4-12b-it-starry-ocean",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Google

Other Google models, starting with the same family.