SSyncDev

AI model comparison

Mistral Medium 3.5 vs Mistral Small 4

Compare pricing, context limits, supported inputs and deployment options to choose a model for your workflow.

Mistral Medium 3.5 and Small 4 are multimodal releases with hosted and downloadable deployment choices. Both publisher pages document a 256K context allowance, but they have different weight terms and token prices.

Hosted rates and licences

The current Medium 3.5 base rates are $1.50 input and $7.50 output per million tokens. Small 4 lists $0.15 input and $0.60 output. These are dated API figures; processing tiers and other conditions can change the bill. Medium’s weights use Modified MIT terms, while Small 4 uses Apache-2.0.

A smaller price does not settle the workflow

Use representative document, screenshot and tool tasks. Validate output formats and check answers against the supplied material. If a route needs repeated attempts or substantial manual repair, a low token rate can still produce an expensive finished result.

Local deployment

Small 4 is an MoE checkpoint with total and active counts that describe different parts of its footprint. Neither an active count nor the hosted rate establishes the full memory requirement. Verify the exact runtime, context and precision for either model.

Side-by-sideCapabilities and practical differences

Mistral AI

Mistral Medium 3.5

Mistral Medium 3.5 is an open-weight multimodal model for coding and agent applications, released.

Context
256K
Max output
262.14K
Input / 1M tokens
$1.50
Output / 1M tokens
$7.50

Accepts

  • Text input
  • Image input

Produces

  • Text output
Full model profiledocs.mistral.ai ↗
VS

Mistral AI

Mistral Small 4

Mistral Small 4 combines instruction-following, reasoning and coding in an Apache-2.0 open-weight model released.

Context
256K
Max output
256K
Input / 1M tokens
$0.15
Output / 1M tokens
$0.60

Accepts

  • Text input
  • Image input

Produces

  • Text output
Full model profiledocs.mistral.ai ↗

At a glance

Key differences

Compare the available figures below, then try both models on the work you need to do.

Mistral Medium 3.5

  • Maximum output: 262.14K

Mistral Small 4

  • Input price: $0.15
  • Output price: $0.60

Context capacity

Higher listed limit

Mistral Medium 3.5:256K

Mistral Small 4:256K

Same value

Maximum output

Higher listed limit

Mistral Medium 3.5:262.14K

Mistral Small 4:256K

Mistral Medium 3.5

Input price

Lower listed token rate

Mistral Medium 3.5:$1.50

Mistral Small 4:$0.15

Mistral Small 4

Output price

Lower listed token rate

Mistral Medium 3.5:$7.50

Mistral Small 4:$0.60

Mistral Small 4

Specifications

Pricing, limits and capabilities

Specifications for Mistral Medium 3.5 and Mistral Small 4. Higher limits and lower prices describe these figures only.
SpecificationMistral Medium 3.5Mistral Small 4
ProviderMistral AIMistral AI
Listed pricing routeMistral APIMistral API
Model familymistral-mediummistral-small
Accepted inputstext, imagetext, image
Produced outputstexttext
Context window256K256K
Max output262.14KHigher256K
Input price / 1M$1.50$0.15Lower
Output price / 1M$7.50$0.60Lower
Reasoning modeDocumentedDocumented
Tool callingDocumentedDocumented
Image inputDocumentedDocumented
Audio inputNot documentedNot documented
Structured outputDocumentedDocumented
LicenseModified MITApache-2.0

Workload guidance

Which model should you choose?

Start with a measured baseline

Test Small 4 for the repeated bounded work in your application. Include Medium 3.5 as another candidate for tasks where the smaller route misses an acceptance requirement. Use the same tools and output budget, and review errors by task rather than one blended average.

Choose from accepted-task cost, response time and correction effort. Keep an example of a failed response with the result. The decision then remains reviewable when a model, host or price schedule changes.

Source notes

This is a documentation-based selection guide, not an independent benchmark. Publisher source 1 · Publisher source 2.

Documentation and model profiles

Explore each model’s features, API requirements and licence terms in the sources below.

Questions people ask

It brings together documented inputs and outputs, token limits, deployment terms and available pricing. When both models have source-linked, verified results on a shared benchmark, those results are shown with their individual sources. The comparison is a starting point for evaluating your own workload.

Compare a different pair

Search by model or provider to explore another comparison.

Open model comparison