Is Mistral Small 3.2 24B an API model?

This page documents downloadable publisher weights. Maviter has not reviewed or asserted a provider-hosted API identity, price or data policy for this record.

Does Maviter host Mistral Small 3.2 24B files?

No. Maviter links to the exact publisher repository revision and does not mirror model binaries or third-party quantisations.

Exact downloadable model / source revision pinned

Mistral Small 3.2 24B

Q: Can Mistral Small 3.2 24B run locally?

The publisher provides discoverable model weights. Local compatibility still depends on the exact runtime, quantisation, context and machine. Maviter classifies this as workstation planning territory, not a guarantee.

mistralai/Mistral-Small-3.2-24B-Instruct-2506

A 24B multilingual instruction model with vision and function-calling examples for capable self-hosted systems.

Open pinned publisher source ↗Calculate local fit

Direct answer

Where Mistral Small 3.2 24B fits.

A 24B instruction model intended for capable local or self-hosted systems.

The pinned Mistral model card includes vision and function-calling examples and lists multilingual coverage.

Identity before opinion

The facts this release can support.

Exact model ID: mistralai/Mistral-Small-3.2-24B-Instruct-2506
Release identity: Mistral Small / 95a6d26c4bfb
Parameters: 24.01B 24,01,13,61,280 recorded
Documented context: 1,31,072 tokens Publisher maximum, not a local-fit promise
Input → output: text + image → text
Weights access: Publicly discoverable files
Listed licence: Apache-2.0 Read the complete publisher terms
API status: Not reviewed No API identity, price or data policy asserted

Use-case fit

Consider it when…

Evaluate multilingual assistants on workstation-class hardware
Prototype vision and function-calling workflows
Test self-hosted instruction quality before selecting an API route

Limits and cautions

Check carefully when…

Workstation-class memory does not guarantee acceptable speed
Function calling requires a compatible serving stack and exact template
Vision and language coverage need domain-specific human evaluation

Transparent local planning

What 4-bit weights may ask of the machine.

These are Maviter calculations at an 8,192-token context—not publisher hardware requirements or proof that a runtime supports this release.

01 / Weight floor11.18 GiB

Parameters × 4 bits, converted to GiB.

02 / Runtime + KV3+ GiB

Runtime reserve and formula-backed KV cache.

03 / Planning floor15 GiB

Calculated working set rounded upward.

04 / Comfortable band24 GiB

A planning band with headroom—not a guarantee.

Before adoption

A three-check evaluation is more useful than a leaderboard.

Run the exact publisher release, on the intended runtime, with inputs from the work it must actually perform.

01Verify the documented serving and tokenizer versions
02Test function-call arguments against a strict schema
03Measure representative multilingual and visual inputs

Evidence ledger

Publisher sources, not copied binaries.

Pinned repository treemistralai/Mistral-Small-3.2-24B-Instruct-250695a6d26c4bfb886c58daf9d3f7332c857cb27b43Open source ↗Pinned model cardCapabilities, licence and usage notesREADME.md at the recorded revisionRead card ↗

The evidence ledger pins an exact publisher revision and monitors it for drift. A source check does not independently reproduce benchmarks or certify production suitability.

Continue comparing

Qwen / 27.78BQwen3.6 27B

A dense 27B Qwen vision-language model positioned for agentic coding, reasoning and multimodal work on high-memory systems.

Open evidence brief →Google / 31.27BGemma 4 31B IT

A large open-weight Gemma instruction model for text-and-image reasoning, coding and multimodal understanding on workstation-class systems.

Open evidence brief →DeepSeek / 14.77BDeepSeek R1 Distill Qwen 14B

A larger distilled reasoning model for local systems with more memory headroom and a need for deeper bounded analysis.

Open evidence brief →

Common questions

Know the boundary before downloading.

Can Mistral Small 3.2 24B run locally?

The publisher provides discoverable weights, but runtime support and performance are machine-specific. Use Maviter's fit calculator, then run one controlled test.

Is this an API model?

This record documents downloadable weights. It does not assert a provider-hosted API ID, current price, rate limit or data policy.

Does Maviter host or redistribute the files?

No. The pinned links return to the publisher. Maviter does not mirror binaries or endorse unaudited third-party quantisations.