autoloops
← Models
Model APILLM

Gemma 4 26B A4B IT

26B MoE · ~4B active · Instruction-tuned

Mixture-of-Experts Gemma 4 — 25.2B total parameters with ~3.8B active per token. Built for high-throughput voice and agent workloads with low TTFT.

Model details

Developed byGoogle DeepMind
Model familyGemma 4
ArchitectureMixture-of-Experts (MoE)
Total parameters25.2B
Active parameters~3.8B per token
Experts8 active / 128 total + 1 shared
Context length256K tokens
ModalitiesText, Image → Text
LicenseApache 2.0

On Autoloops

HardwareH100
APIOpenAI-compatible
RegionUS West
Pricing$0.13 / $0.45 / $0.02 per 1M tokens (input / output / cached)
Model IDgemma-4-26b-a4b-it

Capabilities

  • MoE efficiency — dense-model quality at roughly 4B active compute per token
  • Tuned on Autoloops for low time-to-first-token under concurrent voice calls
  • Instruction-tuned with thinking mode and native function calling
  • Image understanding with variable aspect ratios and resolutions
  • OpenAI-compatible chat completions via Autoloops

OpenAI-compatible chat completion

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["AUTOLOOPS_API_KEY"],
    base_url="https://api.autoloops.ai/v1",
)

stream = client.chat.completions.create(
    model="gemma-4-26b-a4b-it",
    messages=[{"role": "user", "content": "Tell me a fun fact about cats."}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Sources

More models