Qwen3 30B-A3B — Spécifications
| Développeur | Alibaba |
|---|---|
| Type | LLM (architecture MoE) |
| Modalité | Texte → Texte |
| Paramètres | 30 milliards au total / 3 milliards actifs (MoE) |
| Fenêtre de contexte | 128 K |
| Sortie maximale | — |
| Licence | Apache 2.0 (ouverte) |
| Poids ouverts | Oui |
| Publié | 2025 |
| Prix de l’entrée | 0,12 $ / 1 million |
| Prix de la sortie | 0,50 $ / 1 million |
| Fournisseurs d'API | Alibaba, OpenRouter, Ollama |
Exécutez-le localement
| VRAM (4 bits) | ~18 Go |
|---|---|
| GPU minimal requis | RTX 4090 24 Go (quantification Q4) — rapide, 3 milliards actifs |
What is Qwen3 30B-A3B?
Qwen3 30B-A3B is a mixture-of-experts with 30B total parameters but only about 3B active
per token — near-32B quality at a fraction of the compute, and very fast locally. Apache 2.0,
128K context, roughly 18 GB of VRAM at 4-bit on an RTX 4090 24GB.
The speed is the reason to pick it. Because only ~3B parameters are active on each token,
generation runs several times faster than a dense 32B on the same card, while memory use is
only modestly higher. That combination matters specifically for local interactive work —
coding assistants, chat interfaces, anything where you are watching tokens appear — where a
dense 32B on consumer hardware is often just slow enough to be irritating. The trade is that
you pay for 30B of memory to get 3B of speed, so on a 24 GB card the dense Qwen3 32B remains
the better choice if raw quality per gigabyte matters more than latency. Choose 30B-A3B when
responsiveness is the feature.
Qwen3 30B-A3B pricing: API cost per 1M tokens
| Entrée (par million de jetons) | $0.120 |
|---|---|
| Sortie (par million de jetons) | $0.500 |
| Output/input ratio | 4.2× |
| Blended (4:1 in:out) | $0.196 per 1M tokens |
What Qwen3 30B-A3B costs per month
Real monthly spend at a 4:1 input-to-output mix — the ratio a typical chat or RAG workload actually produces.
| Charge de travail | Jetons/mois | Cost / month |
|---|---|---|
| Projet secondaire | 1 million en entrée / 0,25 million en sortie | $0.25 |
| Petite équipe | 20 millions en entrée / 5 millions en sortie | $4.90 |
| Production | 200 millions en entrée / 50 millions en sortie | $49 |
Run your own numbers in the Calculateur de coûts des API IA.
Cheaper alternatives to Qwen3 30B-A3B
| Modèle | Coût combiné par million de dollars | You save |
|---|---|---|
| Mistral 7B ouverte | $0.0220 | 89% cheaper |
| Llama 3.1 8B ouverte | $0.0220 | 89% cheaper |
| Mistral NeMo 12B ouverte | $0.0240 | 88% cheaper |
Self-host or pay the API?
Qwen3 30B-A3B is open-weight, so you can run it yourself. It needs ~18 Go of VRAM at 4-bit (RTX 4090 24GB (Q4) — fast, 3B active). Self-hosting only beats the API once your volume is high enough to keep that hardware busy — the calculateur auto-hébergement vs API works out the break-even point for your token volume.
Questions fréquemment posées
How much does Qwen3 30B-A3B cost per 1M tokens?
Qwen3 30B-A3B costs $0.120 per 1M input tokens and $0.500 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $0.196 per 1M tokens.
How much does Qwen3 30B-A3B cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $4.90 on Qwen3 30B-A3B. A side project (1M in / 0.25M out) costs roughly $0.25.
What is a cheaper alternative to Qwen3 30B-A3B?
Mistral 7B is the strongest cheaper option in our database at $0.0220 per 1M blended — about 89% less than Qwen3 30B-A3B. It is also open-weight, so self-hosting is an option.
Can I run Qwen3 30B-A3B locally?
Yes. Qwen3 30B-A3B is open-weight and needs about ~18 GB of VRAM at 4-bit quantisation (RTX 4090 24GB (Q4) — fast, 3B active).
Why does Qwen3 30B-A3B 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. Qwen3 30B-A3B charges 4.2× 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 Base de données des modèles d'IA ou le Classement des grands modèles linguistiques (LLM).
