NOUMENOpen app
NOUMEN

Google · Open weights

Gemma 4 31B Split-Untied

Added Sep 17, 2026Open weights

Compare models

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

Loading models…

Use in the app

Blazed-Forge's Split-Untied is a text-only Gemma 4 31B community finetune with an untied BF16 output head, built for creative writing, roleplay, expressive dialogue, and tool use.

  • Tools
Input
Text
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.10
Output
$0.30

Reference prices as published by the model provider. API example

Specifications

Context window
262.1K
Max output
32.8K

Side by side

Gemma 4 31B Split-Untied 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 31B Split-UntiedGoogleThis modelGLM 5.3 FlashZ.aiRemove
ProviderGoogleZ.ai
Discovery tierOpen weightsOpen weights
AddedSep 17, 2026Aug 26, 2026
Limits
Context window262.14K1.05MBest
Max output32.77K131.07KBest
Average output—1.3K tokens
Pricing, USD per 1M tokens
Input$0.10$0.10
Output$0.30$0.30
Cache read—$0.025
Performance
Throughput—174.6 tok/s
Time to first token—2.5 s
Request success (24h)99.7%Best99.6%
Benchmark indexes
Intelligence Index—41.8
Coding Index—71.5
Capabilities
ReasoningNoYes
ToolsYesYes
VisionNoYes
Structured outputNoYes
Video inputNoYes
Reasoning effort—low, high, max

Public evaluations

Benchmarks

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

  • Gemma 4 31B Split-Untied
  • GLM 5.3 Flash

Intelligence Index

  • Gemma 4 31B Split-Untied—
  • GLM 5.3 Flash41.8
3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 71.5

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 60.4%

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 25.4%

  • Harvey LAB-AAv1.1

    Legal tasks passed in full, with no material hallucination

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 0.8%

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 1,452 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 1,642 Elo

Document reasoning

  • GDP.pdf

    Professional reasoning over PDF documents, all-pass rate

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 15.4%

  • AA-LCRv1.1

    Reasoning across long documents

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 80.0%

  • MLCR-AA

    Long-context reasoning over medical records

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 11.7%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 39.9%

  • GPQA Diamond

    Graduate-level science questions

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 91.2%

Coding

  • Terminal-Benchv4.0

    Practical software engineering in a terminal

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 32.8%

  • SciCode

    Scientific computing in Python

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 51.6%

Legacy evaluations

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Gemma 4 31B Split-Untied: —
    • GLM 5.3 Flash: 80.0%

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.

Prompt cache hits
0.0%
Request success (24h)
99.7%
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 82.7%
  • 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": "google/gemma4-31b-splituntied",
    "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="google/gemma4-31b-splituntied",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Google

Other Google models, starting with the same family.