Retired Aug 2026
Mistral Medium 3
Mistral AI's mid-tier model, released May 2025, with near-frontier performance and self-hostable on as few as four GPUs.
Key Specifications
- Context window
- 128K tokens
- Max output
- 33K tokens
- Released
- Inputs
- Text, image
- Capabilities Show details
- Function calling Connect to external tools, APIs, and systems. Structured output Return responses in structured formats like JSON.
Frequently Asked Questions
What is Mistral Medium 3 good for?
Mid-tier general work with text and image input. Mistral said it could be deployed on a single multi-GPU node through its own offering.
When is Mistral Medium 3 not a good fit?
Any new work: Mistral retired it from its API in August 2026. Closed weights, so no self-hosting, and its maximum output limits long single responses.
How does Mistral Medium 3 compare with Mistral Medium 3.5?
Mistral calls Mistral Medium 3.5 its first merged flagship after Mistral Medium 3, folding instruction following, reasoning and coding into one dense set of weights with reasoning effort set per request, for long coding and productivity runs. Same text and image input. Mistral Medium 3 came out 11 months earlier, has a smaller context (128K against 256K tokens) and keeps its weights closed, where Mistral Medium 3.5 publishes them under Modified MIT.
Can I self-host Mistral Medium 3?
No. Mistral does not publish the weights; Mistral Medium 3 is available only through hosted APIs.
More from Mistral
| Model | Context | Input /1M | Output /1M |
|---|---|---|---|
|
Ministral 3 14B
Open weights
|
256K | $0.20 | $0.20 |
|
Ministral 3 3B
Open weights
|
256K | $0.10 | $0.10 |
|
Ministral 3 8B
Open weights
|
256K | $0.15 | $0.15 |
|
Mistral Large 3
Open weights
|
256K | $0.50 | $1.50 |
|
Mistral Medium 3.5
Open weights
|
256K | $1.50 | $7.50 |
|
Mistral Small 4
Open weights
|
256K | $0.15 | $0.60 |
|
Codestral
|
128K | $0.30 | $0.90 |
|
Devstral Medium
|
128K | ||