Mistral-7B-Instruct-v0.3
- Reported parameters
- 7.25B
- Release license
- apache-2.0
Downloadable models · Side by side
Mistral 7B Instruct v0.3 vs Llama 3.1 8B Instruct
Mistral 7B Instruct v0.3 and Llama 3.1 8B Instruct are two of the most common small open models to run yourself. Both fit on a single 8 GB GPU at 4-bit. They differ mainly in license and context length.
Llama-3.1-8B-Instruct
- Reported parameters
- 8.03B
- Release license
- llama3.1
The short answer
Choose Mistral 7B v0.3 if you want a plain Apache 2.0 license and slightly lower memory use: about 5.3 GB at 4-bit for one 8K-token conversation. Choose Llama 3.1 8B if you need long context: it supports 128K tokens against Mistral's 32,768, and needs about 5.7 GB at 4-bit.
Compare the details
Compare context, supported inputs and usage costs. Highlighted rows show a difference.
| Feature | Mistral-7B-Instruct-v0.3 | Llama-3.1-8B-Instruct |
|---|---|---|
| Input formats | text | text |
| Output formats | text | text |
| Context / position limit | Unknown | 131,072 |
| Parameter count | 7,248,023,552 | 8,030,261,248 |
| Release license | apache-2.0 | llama3.1 |
| USD input / 1M tokens | Unknown | $0.02 |
| USD output / 1M tokens | Unknown | $0.04 |
| Model identifier | mistralai/Mistral-7B-Instruct-v0.3 | meta-llama/Llama-3.1-8B-Instruct |
API pricing
Prices are in USD per million tokens. Input is what you send; output is what the model generates. Chat subscriptions and API usage are billed separately.
Tools, images and additional reasoning can add charges. The examples below cover input and output tokens only.
Compare local modelsBeyond the price tag
License
Mistral 7B Instruct v0.3 is Apache 2.0: use it commercially with no extra conditions. Llama 3.1 8B uses the Llama 3.1 Community License, which allows commercial use but adds Meta's acceptable use policy and needs a separate license above 700 million monthly users.
Context length
Llama 3.1 8B reads up to 128K tokens. Mistral 7B v0.3 reads up to 32,768. If you feed in long documents or keep long chat histories, Llama avoids having to split them, though long contexts need much more memory.
Memory
For one 8K-token conversation with a 20% allowance, Mistral 7B needs about 5.3 GB at 4-bit, 9.4 GB at 8-bit and 17.5 GB at 16-bit. Llama 3.1 8B needs about 5.7 GB, 10.2 GB and 19.2 GB. Both models use the same context memory per token, so the gap comes from Llama's extra 0.8 billion parameters.
Tool calling
Mistral 7B v0.3 added function-calling support and an extended vocabulary over v0.2. Llama 3.1 8B Instruct also supports tool use in its chat template. Check that your inference server supports the right template for either.
Common questions
How much VRAM do Mistral 7B and Llama 3.1 8B need?
For one 8K-token conversation including a 20% allowance: Mistral 7B needs about 5.3 GB at 4-bit, 9.4 GB at 8-bit and 17.5 GB at 16-bit; Llama 3.1 8B needs about 5.7 GB, 10.2 GB and 19.2 GB. An 8 GB GPU runs either at 4-bit, and a 24 GB GPU runs either at full precision.
Which has the longer context window?
Llama 3.1 8B, with 128K tokens. Mistral 7B Instruct v0.3 supports 32,768 tokens.
Can I use them commercially?
Yes. Mistral 7B v0.3 is Apache 2.0 with no extra conditions. Llama 3.1 allows commercial use under its community license, which adds an acceptable use policy and requires a separate license above 700 million monthly users.
Can I run them on a CPU-only VPS?
Yes, at 4-bit with about 8 GB of RAM for Mistral or 8 GB for Llama, including the operating system. CPU inference is much slower than a GPU, so it suits testing and low-traffic tools.
Run it yourself: memory and hosting
Memory needed to load each model and hold one 8,192-token conversation, with a 20% allowance for runtime buffers. The GPU and server columns show the cheapest listed options in our catalog that fit.
| Model and precision | Weights | Context memory | Plan for | Cheapest GPU that fits |
|---|---|---|---|---|
| Mistral-7B-Instruct-v0.34-bit | 3.4 GB | 1.0 GB | 5.3 GB | NVIDIA RTX A5000 (24 GB) · $0.27/hr Runpod · about $197/mo always on |
| Mistral-7B-Instruct-v0.38-bit | 6.8 GB | 1.0 GB | 9.4 GB | NVIDIA RTX A5000 (24 GB) · $0.27/hr Runpod · about $197/mo always on |
| Mistral-7B-Instruct-v0.316-bit | 13.5 GB | 1.0 GB | 17.5 GB | NVIDIA RTX A5000 (24 GB) · $0.27/hr Runpod · about $197/mo always on |
| Llama-3.1-8B-Instruct4-bit | 3.7 GB | 1.0 GB | 5.7 GB | NVIDIA RTX A5000 (24 GB) · $0.27/hr Runpod · about $197/mo always on |
| Llama-3.1-8B-Instruct8-bit | 7.5 GB | 1.0 GB | 10.2 GB | NVIDIA RTX A5000 (24 GB) · $0.27/hr Runpod · about $197/mo always on |
| Llama-3.1-8B-Instruct16-bit | 15.0 GB | 1.0 GB | 19.2 GB | NVIDIA RTX A5000 (24 GB) · $0.27/hr Runpod · about $197/mo always on |
Mistral-7B-Instruct-v0.3 on a CPU server
At 4-bit, plan for about 7.3 GB of RAM including the operating system. CPU inference is much slower than a GPU, so this suits testing, background jobs and low traffic.
- Cheapest available server that fits: OVHcloud VPS-2 (US), $8.50/mo (8 GB RAM)
- Size it for your context and users
Llama-3.1-8B-Instruct on a CPU server
At 4-bit, plan for about 7.7 GB of RAM including the operating system. CPU inference is much slower than a GPU, so this suits testing, background jobs and low traffic.
- Cheapest available server that fits: OVHcloud VPS-2 (US), $8.50/mo (8 GB RAM)
- Size it for your context and users
Estimates, not measurements. 4-bit and 8-bit figures assume common quantized files (such as GGUF Q4/Q8); real files add a little overhead. Longer contexts and more simultaneous users need more memory.