Qwen3 8B — Specifiche
Compiled by Mustafa Ihsan from the vendor’s published documentation · Last updated
| Sviluppatore | Alibaba |
|---|---|
| Tipo | LLM (densa) |
| Modalità | Testo → Testo |
| Parametri | 8B |
| Finestra contestuale | 128K |
| Licenza | Apache 2.0 (open) |
| Pesi aperti | Sì |
| Pubblicato | 2025 |
| Prezzo dell’input | $0.04 /1M |
| Prezzo dell’output | $0.14 /1M |
| Provider API | Alibaba, OpenRouter, Ollama |
Esegui localmente
| VRAM (4-bit) | ~5 GB |
|---|---|
| GPU minima | RTX 3060 8GB / qualsiasi GPU da 8 GB |
What is Qwen3 8B?
Qwen3 8B is a small, fast dense model from the Qwen3 family — Apache 2.0, a 128K context,
and about 5 GB of VRAM at 4-bit, which fits an RTX 3060 8GB or any 8 GB card. Hosted pricing
is $0.04 in / $0.14 out per million tokens.
Among models that fit an 8 GB GPU, it is one of the strongest available, and the 128K
context is what separates it from the older generation in the same bracket — Mistral 7B, its
closest historical equivalent, tops out at 32K. That difference decides whether a local
assistant can hold a real document or only a few pages. Apache 2.0 licensing means no
conditions to clear with legal, and the small footprint leaves room on the card for a
generous context or a second process. Treat it as the entry point to local inference: it will
run on hardware you almost certainly already have, and if it proves the use case, Qwen3 14B
and 32B are drop-in upgrades within the same family and licence as your hardware
allows.
Qwen3 8B pricing: API cost per 1M tokens
| Input (per ogni milione di token) | $0.0400 |
|---|---|
| Output (per ogni milione di token) | $0.140 |
| Rapporto output/input | 3.5× |
| Combinato (4:1 in:out) | $0.0600 per 1 milione di token |
What Qwen3 8B costs per month
Spesa mensile reale con un mix input-to-output 4:1 — il rapporto effettivamente prodotto da un tipico carico di lavoro basato su chat o RAG.
| Carico di lavoro | Token/mese | Costo/mese |
|---|---|---|
| Progetto secondario | 1 milione in ingresso / 0,25 milioni in uscita | $0.08 |
| Piccolo team | 20 milioni in ingresso / 5 milioni in uscita | $1.50 |
| Produzione | 200 milioni in ingresso / 50 milioni in uscita | $15 |
Calcola i tuoi numeri personalizzati nel Calcolatore dei costi delle API per l'IA.
Cheaper alternatives to Qwen3 8B
| Modello | Costo combinato ($/1 milione) | Risparmi |
|---|---|---|
| Mistral 7B aperta | $0.0220 | 63% più economico |
| Llama 3.1 8B aperta | $0.0220 | 63% più economico |
| Mistral NeMo 12B aperta | $0.0240 | 60% cheaper |
Eseguirlo in autonomia o pagare l’API?
Qwen3 8B is open-weight, so you can run it yourself. It needs ~5 GB of VRAM at 4-bit (RTX 3060 8GB / any 8GB GPU). Self-hosting only beats the API once your volume is high enough to keep that hardware busy — the calcolatore self-hosting vs API calcola il punto di pareggio in base al tuo volume di token.
Domande frequenti
How much does Qwen3 8B cost per 1M tokens?
Qwen3 8B costs $0.0400 per 1M input tokens and $0.140 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $0.0600 per 1M tokens.
How much does Qwen3 8B cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $1.50 on Qwen3 8B. A side project (1M in / 0.25M out) costs roughly $0.08.
What is a cheaper alternative to Qwen3 8B?
Mistral 7B is the strongest cheaper option in our database at $0.0220 per 1M blended — about 63% less than Qwen3 8B. It is also open-weight, so self-hosting is an option.
Can I run Qwen3 8B locally?
Yes. Qwen3 8B is open-weight and needs about ~5 GB of VRAM at 4-bit quantisation (RTX 3060 8GB / any 8GB GPU).
Why does Qwen3 8B 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 8B charges 3.5× more for output, which is why prompt-heavy workloads are far cheaper to run than generation-heavy ones.
I prezzi indicati sono le tariffe ufficiali pubblicate per l'API principale del modello e vengono aggiornati man mano che i fornitori li modificano. Sconti per volumi elevati, elaborazione batch e input memorizzati nella cache non sono inclusi. Confronta tutti i modelli fianco a fianco nel Database di modelli IA o il Classifica LLM.