Ministral 3 8B — Specifications
Compiled by Mustafa Ihsan from the vendor’s published documentation · Last updated
| Developer | Mistral AI |
|---|---|
| Type | LLM (dense, multimodal) |
| Modality | Text, Image → Text |
| Parameters | 8.4B + 0.4B vision encoder |
| Context window | 256K |
| License | Apache 2.0 (open) |
| Open weights | Yes |
| Released | 2025-12 |
| Input price | $0.15 /1M |
| Output price | $0.15 /1M |
| API providers | Mistral API, Amazon Bedrock, OpenRouter, Ollama (local) |
Run it locally
| VRAM (4-bit) | ~8 GB |
|---|
What is Ministral 3 8B?
Ministral 3 8B is the middle model of Mistral AI’s Ministral 3 family, released on 2 December 2025 alongside Mistral Large 3 under the Apache 2.0 licence. It pairs an 8.4B language model with a 0.4B vision encoder, so it reads images as well as text, and it has a 256K-token context window. Mistral built the family for edge and local use and ships base, instruct and reasoning versions (per Mistral’s launch post).
On Mistral’s API (ministral-8b-2512) it costs $0.15 per million input tokens and $0.15 per million output tokens, with cached input at $0.015 (per Mistral’s pricing page). Amazon Bedrock and OpenRouter also carry it. For local use, Mistral’s model page puts the approximate minimum at 8 GB of GPU memory for 4-bit weights with a 16K context, 17 GB at 4-bit with a quarter of the full context, and 67 GB for BF16 at the full 256K. Mistral’s own GGUF release has a 5.2 GB Q4_K_M file plus a 0.86 GB vision projector, and Ollama’s default ministral-3:8b build is a 6.0 GB download.
Mistral names it the replacement for both Mistral 7B (retired from Mistral’s API on 30 March 2025) and Mistral NeMo 12B (retired on 31 July 2026). Compared with Mistral 7B it adds image input and a 256K window instead of 32K. For other 8B-class options see Llama 3.1 8B and Qwen3 8B, and check what fits your GPU with the LLM VRAM calculator.
Ministral 3 8B pricing: API cost per 1M tokens
| Input (per 1M tokens) | $0.150 |
|---|---|
| Output (per 1M tokens) | $0.150 |
| Output/input ratio | 1× |
| Blended (4:1 in:out) | $0.150 per 1M tokens |
What Ministral 3 8B costs per month
Real monthly spend at a 4:1 input-to-output mix — the ratio a typical chat or RAG workload actually produces.
| Workload | Tokens / month | Cost / month |
|---|---|---|
| Side project | 1M in / 0.25M out | $0.19 |
| Small team | 20M in / 5M out | $3.75 |
| Production | 200M in / 50M out | $38 |
Run your own numbers in the AI API cost calculator.
Cheaper alternatives to Ministral 3 8B
| Model | Blended $/1M | You save |
|---|---|---|
| Llama 3.1 8B open | $0.0240 | 84% cheaper |
| Mistral NeMo 12B open | $0.0292 | 81% cheaper |
| Gemma 3 4B open | $0.0600 | 60% cheaper |
Self-host or pay the API?
Ministral 3 8B is open-weight, so you can run it yourself. It needs ~8 GB of VRAM at 4-bit. Self-hosting only beats the API once your volume is high enough to keep that hardware busy — the self-hosting vs API calculator works out the break-even point for your token volume.
Frequently asked questions
How much does Ministral 3 8B cost per 1M tokens?
Ministral 3 8B costs $0.150 per 1M input tokens and $0.150 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $0.150 per 1M tokens.
How much does Ministral 3 8B cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $3.75 on Ministral 3 8B. A side project (1M in / 0.25M out) costs roughly $0.19.
What is a cheaper alternative to Ministral 3 8B?
Llama 3.1 8B is the strongest cheaper option in our database at $0.0240 per 1M blended — about 84% less than Ministral 3 8B. It is also open-weight, so self-hosting is an option.
Can I run Ministral 3 8B locally?
Yes. Ministral 3 8B is open-weight and needs about ~8 GB of VRAM at 4-bit quantisation.
Why does Ministral 3 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. Ministral 3 8B charges 1× 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 AI models database or the LLM leaderboard.