Mistral Medium 3.5
Mistral Medium 3.5 is a dense 128B-parameter model released in April 2026 with open weights under a modified MIT license. It handles text and images, has a 256K context window and lets you switch reasoning on or off per request. Mistral uses it to power the remote agents in its Vibe agent product.
Specifications
- Context window
- 256Kabout 197,000 words, or 390 pages
- Max output
- Not published
- Input price
- $1.50per 1M tokens
- Output price
- $7.50per 1M tokens
- Cached input
- Not published
- Weights
- OpenModified MIT
- Parameters
- 127.7Bdense
- Input types
- Text, Images
- Output types
- Text
- Reasoning mode
- Yes
- Tool calling
- Yes
- Released
- Apr 28, 2026Current
Mistral API price (mistral-medium-3-5). The docs page shows $1.5 and $7.5 per 1M tokens without input/output labels; mapped in Mistral's usual input-then-output order.
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 | $180 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $48 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $135 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- Per-request reasoning switch (reasoning_effort none or high)
- Native function calling and JSON output
- Dense architecture makes memory planning simple
Trade-offs to weigh
- License withdraws rights from companies above $20M monthly revenue unless they get a commercial license
- 128B dense weights need a multi-GPU server (8-way tensor parallel in examples)
Where it fits
Mistral AI positions Mistral Medium 3.5 for:
Where you can use it
- Hugging Face
- Ollama
- API
Run it yourself
Released under Modified MIT. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 255 GB |
| 8-bit (FP8/INT8) | 128 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 77 GB |
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 | |
|---|---|---|---|---|
| Claude Sonnet 5.5Anthropic | Advanced | 1M | $2 / $10 | Compare |
| GPT-6.1 SolOpenAI | Advanced | 1.05M | $2 / $10 | Compare |
| Muse Spark 1.3Meta | Advanced | 1M | $1.25 / $4.25 | Compare |
| Gemini 3.8 FlashGoogle | Advanced | 1M | $0.75 / $3.75 | Compare |
Frequently asked questions
How much does Mistral Medium 3.5 cost?
Mistral Medium 3.5 lists at $1.50 per million input tokens and $7.50 per million output tokens (checked October 10, 2026). Mistral API price (mistral-medium-3-5). The docs page shows $1.5 and $7.5 per 1M tokens without input/output labels; mapped in Mistral's usual input-then-output order. Confirm current rates on the provider's pricing page before you commit.
What is the context window of Mistral Medium 3.5?
Mistral Medium 3.5 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 Medium 3.5 open source?
Mistral Medium 3.5 is an open-weight model released under the Modified MIT license. You can download the weights from Hugging Face (mistralai/Mistral-Medium-3.5-128B) 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 Medium 3.5 locally?
Yes, if you have enough memory. The weights alone need about 77 GB at 4-bit quantization and about 255 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 Medium 3.5 best for?
Mistral AI positions Mistral Medium 3.5 for coding, agents, enterprise, multimodal. 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-medium-3-5-26-04
- huggingface.co/mistralai/Mistral-Medium-3.5-128B
- huggingface.co/mistralai/Mistral-Medium-3.5-128B/raw/main/config.json
- huggingface.co/mistralai/Mistral-Medium-3.5-128B/raw/main/LICENSE
- huggingface.co/api/models/mistralai/Mistral-Medium-3.5-128B
- 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.