Mistral Models: Efficient Local AI
Mistral AI from Paris builds some of the most efficient open models available. Their 7B model punches well above its weight, and Codestral is purpose-built for coding tasks. If you want maximum quality from limited RAM, Mistral models are a top pick.
- developer
- Mistral AI
- local models
- 14
- API-only models
- 2
- size range
- 3.4B-123B
- RAM range
- 6-96 GB
All Mistral models
14 local models| Model | Size | Quant | VRAM | Min RAM | Best For | Quality | Ollama |
|---|---|---|---|---|---|---|---|
| Ministral 3 3B | 3.4B | Q4_K_M | 3 GB | 6 GB | Chat, Vision, Edge | 74 | |
| Mistral 7B Instruct | 7B | Q4_K_M | 5.5 GB | 8 GB | Chat, Coding | 70 | |
| Mistral 7B Instruct (Q5) | 7B | Q5_K_M | 4.8 GB | 8 GB | Chat, Coding | 71 | |
| Mistral 7B Instruct (Q8) | 7B | Q8_0 | 7.2 GB | 16 GB | Chat, Coding | 72 | |
| Ministral 3 8B | 8.4B | Q4_K_M | 6 GB | 12 GB | Chat, Coding, Vision | 85 | |
| Mistral Nemo 12B | 12B | Q4_K_M | 9.5 GB | 16 GB | Chat, Translation | 82 | |
| Mistral Nemo 12B (Q8) | 12B | Q8_0 | 12.1 GB | 24 GB | Chat, Translation | 84 | |
| Ministral 3 14B | 13.5B | Q4_K_M | 9.1 GB | 16 GB | Chat, Reasoning, Vision | 88 | |
| Mistral Small 22B | 22B | Q4_K_M | 17 GB | 32 GB | Coding, Quality | 84 | |
| Mistral Small 3.1 | 24B | Q4_K_M | 15 GB | 24 GB | Chat, Coding | 86 | |
| Mistral Small 3.1 (Q8) | 24B | Q8_0 | 23.3 GB | 36 GB | Chat, Coding | 88 | |
| Devstral Small 2 24B | 24B | Q4_K_M | 15.2 GB | 24 GB | Agentic coding, Software engineering | 91 | |
| Mixtral 8x7B Instruct | 46.7B | Q4_K_M | 30 GB | 48 GB | Coding, Quality | 76 | |
| Devstral 2 123B | 123B | Q4_K_M | 74.9 GB | 96 GB | Agentic coding, Software engineering | 94 |
Key features
- Excellent performance-per-parameter ratio
- Sliding window attention for efficiency
- Strong instruction following
- Codestral specialized for coding
Cloud / API-only models
2 APIWeights are open but far beyond consumer hardware. These run as APIs. Each page lists the local alternatives.
| Model | Size | Access | Best For |
|---|---|---|---|
| Mistral Medium 3.5 | 128B | API only | Coding, Quality, Tool calling |
| Mistral Large 4 | 1049B | API only | Reasoning, Agentic coding, Vision, Long context |
Device compatibility
Which Mistral models can run on each device class, based on minimum RAM requirements.
| Model | iPhone | Air | Pro | Studio | Mini | |
|---|---|---|---|---|---|---|
| Ministral 3 3B (3.4B) | Possible | Possible | Excellent | Excellent | Excellent | Possible |
| Mistral 7B Instruct (7B) | Possible | Possible | Excellent | Excellent | Excellent | Possible |
| Mistral 7B Instruct (Q5) (7B) | Possible | Possible | Excellent | Excellent | Excellent | Possible |
| Mistral 7B Instruct (Q8) (7B) | No | Possible | Possible | Excellent | Possible | Possible |
| Ministral 3 8B (8.4B) | Possible | Possible | Possible | Excellent | Possible | Possible |
| Mistral Nemo 12B (12B) | No | Possible | Possible | Excellent | Possible | Possible |
| Mistral Nemo 12B (Q8) (12B) | No | Possible | Possible | Possible | Possible | Possible |
| Ministral 3 14B (13.5B) | No | Possible | Possible | Excellent | Possible | Possible |
| Mistral Small 22B (22B) | No | Possible | Possible | Possible | Possible | Possible |
| Mistral Small 3.1 (24B) | No | Possible | Possible | Possible | Possible | Possible |
| Mistral Small 3.1 (Q8) (24B) | No | No | Possible | Possible | Possible | No |
| Devstral Small 2 24B (24B) | No | Possible | Possible | Possible | Possible | Possible |
| Mixtral 8x7B Instruct (46.7B) | No | No | Possible | Possible | Possible | No |
| Devstral 2 123B (123B) | No | No | Possible | Possible | No | No |
RAM requirements
Largest dense Q4 model that fits each memory tier
Get told when a better model fits your MacBook Air M5 16 GB
Get one email when a new open-weight model is a better fit for this machine.
For: MacBook Air M5 16 GB
You'll also get the weekly ModelFit email. By signing up you agree to our Privacy Policy. Unsubscribe from either email anytime, in one click.
Config reports
No speed test run for this family yet. Run the speed test: npx @modelfit/cli@latest bench
Share your config
Frequently asked questions
What is the best Mistral model for 16GB RAM?
Mistral Nemo 12B Q4 is the best fit, using about 8.5GB. If you want something lighter, Mistral 7B Q4 uses 5.5GB and still delivers strong results.
Is Codestral worth using for coding?
Yes. Codestral 22B is specifically trained for code generation and performs better than general models of similar size. It needs 20GB RAM (Q4).
How does Mistral compare to Llama at 7B?
Very close in benchmarks. Mistral 7B uses sliding window attention for better efficiency with long contexts. Llama 3.1 8B has a slight edge on reasoning tasks. Both are excellent choices.
