Gemma 3 27B — Specifiche
| Sviluppatore | |
|---|---|
| Tipo | LLM (multimodale) |
| Modalità | Testo, immagine → testo |
| Parametri | 27B |
| Finestra contestuale | 128K |
| Output massimo | — |
| Licenza | Gemma (open) |
| Pesi aperti | Sì |
| Pubblicato | 2025 |
| Prezzo dell’input | 0,08 $ /1M |
| Prezzo dell’output | 0,16 $ / 1 milione |
| Provider API | Google AI Studio, OpenRouter, Ollama |
Esegui localmente
| VRAM (4-bit) | ~16 GB |
|---|---|
| GPU minima richiesta | RTX 4080 16 GB / RTX 4090 |
What is Gemma 3 27B?
Gemma 3 27B is the largest open model in Google’s Gemma 3 line — multimodal across text
and images, 128K context, 140+ languages — and the strongest single-GPU local model in its
class. It needs roughly 16 GB of VRAM at 4-bit, so an RTX 4080 16GB or an RTX 4090.
The phrase “single-GPU” is what makes it interesting. Above this size, open models tend to
jump straight to multi-GPU servers: 70B-class models want 40 GB, and the large
mixture-of-experts flagships want hundreds. Gemma 3 27B is close to the ceiling of what one
consumer card can hold while remaining genuinely capable, which makes it the practical
maximum for a workstation deployment. Teams that need on-premises inference for compliance
reasons, and can live with 27B-class quality rather than frontier quality, can serve it from
a single machine instead of a rack. The API price of $0.08 in / $0.16 out per million tokens
is low enough that self-hosting only pays at sustained volume — worth checking against the
self-hosting vs API
calcolatore before buying hardware.
Gemma 3 27B pricing: API cost per 1M tokens
| Input (per ogni milione di token) | $0.0800 |
|---|---|
| Output (per ogni milione di token) | $0.160 |
| Output/input ratio | 2× |
| Blended (4:1 in:out) | $0.0960 per 1M tokens |
What Gemma 3 27B costs per month
Real monthly spend at a 4:1 input-to-output mix — the ratio a typical chat or RAG workload actually produces.
| Carico di lavoro | Token/mese | Cost / month |
|---|---|---|
| Progetto secondario | 1 milione in ingresso / 0,25 milioni in uscita | $0.12 |
| Piccolo team | 20 milioni in ingresso / 5 milioni in uscita | $2.40 |
| Produzione | 200 milioni in ingresso / 50 milioni in uscita | $24 |
Run your own numbers in the Calcolatore dei costi delle API per l'IA.
Cheaper alternatives to Gemma 3 27B
| Modello | Costo combinato ($/1 milione) | You save |
|---|---|---|
| Phi-4 aperta | $0.0840 | 13% cheaper |
Self-host or pay the API?
Gemma 3 27B is open-weight, so you can run it yourself. It needs ~16 GB of VRAM at 4-bit (RTX 4080 16GB / RTX 4090). Self-hosting only beats the API once your volume is high enough to keep that hardware busy — the calcolatore self-hosting vs API works out the break-even point for your token volume.
Domande frequenti
How much does Gemma 3 27B cost per 1M tokens?
Gemma 3 27B costs $0.0800 per 1M input tokens and $0.160 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $0.0960 per 1M tokens.
How much does Gemma 3 27B cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $2.40 on Gemma 3 27B. A side project (1M in / 0.25M out) costs roughly $0.12.
What is a cheaper alternative to Gemma 3 27B?
Phi-4 is the strongest cheaper option in our database at $0.0840 per 1M blended — about 13% less than Gemma 3 27B. It is also open-weight, so self-hosting is an option.
Can I run Gemma 3 27B locally?
Yes. Gemma 3 27B is open-weight and needs about ~16 GB of VRAM at 4-bit quantisation (RTX 4080 16GB / RTX 4090).
Why does Gemma 3 27B charge more for output than input?
Output tokens are generated one at a time and cannot be batched the way a prompt can, so they cost the provider more to serve. Gemma 3 27B charges 2× more for output, which is why prompt-heavy workloads are far cheaper to run than generation-heavy ones.
Prices are the published list rates for the model's primary API and are reviewed as providers change them. Volume, batch and cached-input discounts are not included. Compare every model side by side in the Database di modelli di intelligenza artificiale o il Classifica LLM.
