plubo checking

Small models,
served well.

Three small open models on an OpenAI-compatible API. All of them take text, images, audio and video in. Point your existing client at us and change the model name.

The lineup

Gemma 4 12Bgemma-4-12b-it

The flagship, and the one to reach for when the answer matters more than the price. Full 262K context.

262,144 context text · image · audio · video $0.06 / $0.30 per 1M

Gemma 4 E4Bgemma-4-e4b-it

Most of the quality for a little over half the price. The sensible default for ordinary work.

131,072 context text · image · audio · video $0.04 / $0.20 per 1M

Gemma 4 E2Bgemma-4-e2b-it

The smallest one. For classification, extraction and routing at volume, where throughput and cost are the whole question.

131,072 context text · image · audio · video $0.02 / $0.10 per 1M

All three are natively multimodal: audio, images and video go into the model directly, with no transcription step in front and no separate vision endpoint beside it. All three do tool calling and structured outputs. Audio is the part that is hard to find anywhere else — the only Gemma 4 sizes that accept it are the three we run.

Try them

Say something, drop an image, or hold a thought out loud.
No key needed here.

Enjoying it? The API is the same thing, without the rate limit. Ask for a key.

Use them

Point any OpenAI-compatible client at api.plubo-ai.com/v1 and pick a model by name. Nothing else changes.

curl https://api.plubo-ai.com/v1/chat/completions \
  -H "Authorization: Bearer $PLUBO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-4-e4b-it",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Audio goes in directly. Hand any of them a WAV and skip the transcription step.

import base64
audio = base64.b64encode(open("memo.wav", "rb").read()).decode()

client.chat.completions.create(
    model="gemma-4-12b-it",
    messages=[{"role": "user", "content": [
        {"type": "text", "text": "Summarize this in a line."},
        {"type": "input_audio",
         "input_audio": {"data": audio, "format": "wav"}},
    ]}],
)

What they cost

modelin / out per 1M
Gemma 4 12Bgemma-4-12b-it$0.06 / $0.30
Gemma 4 E4Bgemma-4-e4b-it$0.04 / $0.20
Gemma 4 E2Bgemma-4-e2b-it$0.02 / $0.10

Billed per token, in and out, with no minimums. The 12B has no separate vision or audio encoder — media is projected straight into the model — so images, audio and video all bill at its plain input rate. E4B and E2B carry real encoders, so image and audio input is priced separately there. Video bills at the input rate on all three.

You could run these yourself.

They are small enough to fit on your own hardware, which is much of the appeal. If you would rather not run them, they are already running here.

Ask for a key →