Mistral Small 4
Mistral Small 4, released in March 2026, merges Mistral's reasoning, vision and agentic-coding lines into one Apache 2.0 model. It is a 128-expert mixture-of-experts with 119B total and about 6.5B active parameters, and reasoning can be switched on per request. The context window is 256K tokens.
Specifications
- Context window
- 256Kabout 197,000 words, or 390 pages
- Max output
- Not published
- Input price
- $0.15per 1M tokens
- Output price
- $0.60per 1M tokens
- Cached input
- Not published
- Weights
- OpenApache-2.0
- Parameters
- 119.4B total, 6.5B activeMoE
- Input types
- Text, Images
- Output types
- Text
- Reasoning mode
- Yes
- Tool calling
- Yes
- Released
- Mar 16, 2026Current
Mistral API price (mistral-small-2603 / mistral-small-latest) from Mistral docs and launch post.
What it costs in practice
| Workload | Monthly cost at list price |
|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $16 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $4.56 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $13 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- Apache 2.0 license
- Low active-parameter count keeps API prices low
- One model covers chat, reasoning, coding and image input
Trade-offs to weigh
- All 119B parameters must be loaded (about 240 GB at BF16)
- Not in the Ollama library as of October 2026
Where it fits
Mistral AI positions Mistral Small 4 for:
Where you can use it
- Hugging Face
- API
- build.nvidia.com
Run it yourself
Released under Apache-2.0. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 239 GB |
| 8-bit (FP8/INT8) | 119 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 72 GB |
This is a mixture-of-experts model: all 119.4B total parameters must sit in memory, but only about 6.5B are used per token, so it runs faster than a dense model of the same size.
Add memory for the KV cache (it grows with context length and concurrent users) and 1-2 GB of runtime overhead. Size it precisely in the VRAM calculator.
Alternatives to consider
| Model | Tier | Context | Input / output per 1M | |
|---|---|---|---|---|
| Gemma 4 31BGoogle | Efficient | 256K | $0.20 / $0.40 | Compare |
| Granite 4.2 30BIBM | Efficient | 128K | $0.16 / $0.65 | Compare |
| GLM-5.3-FlashZ.ai (Zhipu AI) | Efficient | 1M | $0.15 / $0.50 | Compare |
| Claude Haiku 5.5Anthropic | Efficient | 1M | $0.10 / $0.50 | Compare |
Frequently asked questions
How much does Mistral Small 4 cost?
Mistral Small 4 lists at $0.15 per million input tokens and $0.60 per million output tokens (checked October 10, 2026). Mistral API price (mistral-small-2603 / mistral-small-latest) from Mistral docs and launch post. Confirm current rates on the provider's pricing page before you commit.
What is the context window of Mistral Small 4?
Mistral Small 4 accepts up to 262,144 tokens of context (about 197,000 words, or 390 pages). Long prompts cost more and can be slower, so most apps send only what the task needs.
Is Mistral Small 4 open source?
Mistral Small 4 is an open-weight model released under the Apache-2.0 license. You can download the weights from Hugging Face (mistralai/Mistral-Small-4-119B-2603) and run or fine-tune them yourself, within the license terms. Open weights isn't the same as fully open source: training data and code may not be published.
Can I run Mistral Small 4 locally?
Yes, if you have enough memory. The weights alone need about 72 GB at 4-bit quantization and about 239 GB at 16-bit, plus room for the KV cache. Use the VRAM calculator on LLModel.com to check a specific GPU or Mac.
What is Mistral Small 4 best for?
Mistral AI positions Mistral Small 4 for coding, agents, high volume, self-hosting. It accepts text, images as input, has a reasoning (thinking) mode and supports tool calling. Test it on a sample of your own tasks before you choose it for production.
Sources
- docs.mistral.ai/models/mistral-small-4-0-26-03
- mistral.ai/news/mistral-small-4
- huggingface.co/mistralai/Mistral-Small-4-119B-2603
- huggingface.co/mistralai/Mistral-Small-4-119B-2603/raw/main/config.json
- huggingface.co/api/models/mistralai/Mistral-Small-4-119B-2603
- ollama.com/library
Facts checked against these pages on October 10, 2026. Spot an error? Tell us.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.