Text GenerationRetired
Mistral-Small-3.2-24B-Instruct-2506
- Function Calling
- Streaming
- Structured Output
- Vision
0.09€/Mtoken (input)
0.28€/Mtoken (output)
Licence : Apache 2.0
Parameters : 24B
Quantization : bf16
Max. context : 128K
AI Endpoints: catalogue and playground
Compare and test open-weight models for text, code, voice and vision, no sign-up required. Then move to production with the same OpenAI-compatible API.
API URL
https://oai.endpoints.kepler.ai.cloud.ovh.net/v1
Capabilities
What a model can do beyond its core task. Select several to find models that do all of them.
Can call tools and functions you define, to fetch data or trigger actions.
Works through a problem step by step before answering. Slower, but stronger on complex tasks such as maths, code or planning.
Sends its answer token by token as it is generated, so you can display it right away.
Answers in valid JSON that follows a schema you provide.
Understands images as input, alongside text.
ContextAny
The most text a model can handle in one request, counting the prompt, the conversation history and the answer together, in tokens (roughly ¾ of a word each). Larger windows suit long documents and long conversations. Audio, image and embedding models have none, so narrowing this hides them.
0
6
4K16K32K64K128K256K1M+
SizeAny
Smaller models are generally cheaper and faster, but less capable. Choose a larger model for complex tasks that need state-of-the-art quality. Counted in billions of parameters (B).
0
7
1B3B8B15B30B70B150B400B+
Series
The model family, usually from a single publisher. Models of one series share an architecture and behave alike.
BGE (BAAI General Embedding): embedding models from the Beijing Academy of Artificial Intelligence.
OpenAI's open-weight reasoning models, under Apache 2.0.
Meta's open-weight Llama language models.
Open-weight language models from Mistral AI.
NVIDIA's speech models, here text-to-speech voices by language.
Alibaba Cloud's Qwen models, covering chat, code, vision and embeddings.
Qwen3Guard: Alibaba Cloud's safety models, which flag harmful prompts and answers.
Stability AI's image generation models.
OpenAI's open-source speech recognition models.
Status
Available: ready to use. New: among the latest models added to the catalogue. Beta: in testing, so its behavior and pricing may still change. Being retired: still answering, but scheduled for end of life, so choose another model for new work. Retired: past its end of life and no longer served. It stays listed for a while so you can see where it went.
Use cases
What the publisher recommends the model for.
Tells the speakers in a recording apart and labels who said what.
Classifies prompts and answers as safe or unsafe, to guard an application against harmful content.
Offers several voices to choose from.
Embeds text in many languages in one shared space, so a query in one language matches documents in another.
Transcribes speech and translates it into English.