NOUMENOpen app
NOUMEN

Google · Open weights

Gemma 4 12B Instruct

Added Aug 1, 2026Open weights

Compare models

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

Loading models…

Use in the app

Google's Gemma 4 12B Instruct is an open-weight multimodal model for text and image understanding, with tool calling and structured output support.

  • Tools
  • Vision
  • Structured output
Input
Text, Images
Output
Text
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
32.8K
Average output (7 days)
162 tokens
Longer than 16% of 468 models

Side by side

Gemma 4 12B Instruct 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 InstructGoogleThis modelGemma 4 26B A4B ThinkingGoogleRemove
ProviderGoogleGoogle
Discovery tierOpen weightsOpen weights
AddedAug 1, 2026Apr 2, 2026
Limits
Context window131.07K262.14KBest
Max output32.77K131.07KBest
Average output162 tokens1.1K tokens
Pricing, USD per 1M tokens
Input$0.05Best$0.09
Output$0.25Best$0.30
Cache read$0.025Best$0.05
Performance
Throughput23.5 tok/s31.8 tok/sBest
Time to first token2.5 sBest4.4 s
Request success (24h)98.6%100.0%Best
Benchmark indexes
Intelligence Index9.416.7Best
Coding Index—39.3
Capabilities
ReasoningNoYes
ToolsYesYes
VisionYesYes
Structured outputYesYes
Video inputNoYes
Reasoning effort——

Public evaluations

Benchmarks

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

  • Gemma 4 12B Instruct
  • Gemma 4 26B A4B Thinking

Intelligence Index

  • Gemma 4 12B Instruct9.4
  • Gemma 4 26B A4B Thinking16.7

Gemma 4 12B Instruct: Better than 19% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • Gemma 4 12B Instruct: —
    • Gemma 4 26B A4B Thinking: 39.3

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Gemma 4 12B Instruct: —
    • Gemma 4 26B A4B Thinking: 1.9%

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Gemma 4 12B Instruct: —
    • Gemma 4 26B A4B Thinking: 0.2%

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Gemma 4 12B Instruct: —
    • Gemma 4 26B A4B Thinking: 315 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Gemma 4 12B Instruct: —
    • Gemma 4 26B A4B Thinking: 565 Elo

Document reasoning

  • AA-LCRv1.1

    Reasoning across long documents

    • Gemma 4 12B Instruct: 35.0%
    • Gemma 4 26B A4B Thinking: 65.7%Top

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Gemma 4 12B Instruct: 6.3%
    • Gemma 4 26B A4B Thinking: 19.3%Top

  • GPQA Diamond

    Graduate-level science questions

    • Gemma 4 12B Instruct: 66.1%
    • Gemma 4 26B A4B Thinking: 79.2%Top

  • IFBench

    Following precise output instructions

    • Gemma 4 12B Instruct: 45.2%
    • Gemma 4 26B A4B Thinking: 72.4%Top

Coding

  • SciCode

    Scientific computing in Python

    • Gemma 4 12B Instruct: —
    • Gemma 4 26B A4B Thinking: 40.0%

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • Gemma 4 12B Instruct: 11.4%
    • Gemma 4 26B A4B Thinking: 13.6%Top

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • Gemma 4 12B Instruct: 31.9%
    • Gemma 4 26B A4B Thinking: 43.6%Top

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Gemma 4 12B Instruct: 35.0%
    • Gemma 4 26B A4B Thinking: 65.7%Top

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
23.5 tok/s
Faster than 15% of 254 models
Time to first token
2.5 s
Quicker than 67% of 254 models
Prompt cache hits
30.2%
About $0.0425 per 1M input tokens with caching
Request success (24h)
98.5%
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 5.5%
  • 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",
    "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",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Google

Other Google models, starting with the same family.