NOUMENOpen app
NOUMEN

Google · Open weights

DiffusionGemma

Added Sep 19, 2026Open weights

Compare models

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

Loading models…

Use in the app

DiffusionGemma is a high-speed diffusion-based version of Gemma 4 26B A4B. It supports optional reasoning and a 262,144-token context window. Temperature, min-p, seed, minimum output tokens, logit bias, bad words, and allowed token IDs are unavailable; omit these parameters to use native diffusion sampling.

  • Reasoning
Input
Text
Output
Text
Reasoning effort
none · xhigh
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.05
Output
$0.15
Cache read
$0.025

Reference prices as published by the model provider. API example

Specifications

Context window
262.1K
Max output
32.8K
Average output (7 days)
266 tokens
Longer than 28% of 473 models

Side by side

DiffusionGemma 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…

DiffusionGemmaGoogleThis modelQwen3 Coder FlashQwenRemove
ProviderGoogleQwen
Discovery tierOpen weightsEfficient
AddedSep 19, 2026Sep 17, 2025
Limits
Context window262.14KBest128K
Max output32.77K65.54KBest
Average output266 tokens—
Pricing, USD per 1M tokens
Input$0.05Best$0.30
Output$0.15Best$1.50
Cache read$0.025—
Benchmark indexes
Intelligence Index9.59.6Best
Coding Index19.7—
Math Index—29.0
Capabilities
ReasoningYesNo
Reasoning effortnone, xhigh—

Public evaluations

Benchmarks

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

  • DiffusionGemma
  • Qwen3 Coder Flash

Intelligence Index

  • DiffusionGemma9.5
  • Qwen3 Coder Flash9.6

DiffusionGemma: Better than 19% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • DiffusionGemma: 19.7
    • Qwen3 Coder Flash: —

  • Math Index

    Composite of the math evaluations

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 29.0

Agentic work

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • DiffusionGemma: 321 Elo
    • Qwen3 Coder Flash: —

Document reasoning

  • AA-LCRv1.1

    Reasoning across long documents

    • DiffusionGemma: 19.7%
    • Qwen3 Coder Flash: 32.7%Top

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • DiffusionGemma: 10.8%Top
    • Qwen3 Coder Flash: 3.8%

  • GPQA Diamond

    Graduate-level science questions

    • DiffusionGemma: 66.9%Top
    • Qwen3 Coder Flash: 51.6%

  • MMLU-Pro

    Multi-subject knowledge and reasoning

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 70.6%

  • IFBench

    Following precise output instructions

    • DiffusionGemma: 59.5%Top
    • Qwen3 Coder Flash: 32.7%

Coding

  • LiveCodeBench

    Recent competitive programming problems

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 40.3%

Math

  • AIME 2025

    Competition mathematics, 2025 problems

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 29.0%

  • AIME

    Competition mathematics

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 29.7%

  • MATH-500

    Competition math problems

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 89.3%

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 15.2%

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • DiffusionGemma: —
    • Qwen3 Coder Flash: 34.5%

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • DiffusionGemma: 19.7%
    • Qwen3 Coder Flash: 32.7%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.

Prompt cache hits
0.0%
About $0.05 per 1M input tokens with caching

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.

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/diffusiongemma",
    "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="google/diffusiongemma",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Google

Other Google models, starting with the same family.