Mistral Medium 3.5 (batch)

Overview

Mistral Medium 3.5 (batch) is Mistral’s dense 128B instruction-following model, identified as mistralai/mistral-medium-3-5:batch. It accepts text, images and files, and produces text. Its stated focus is agentic workflows, coding and complex tasks, making its broad input support relevant to teams building assistants that need to work across more than text alone.

The model has a 262,144-token context window and supports reasoning and tools. Its maximum output is 209,715 tokens. These limits give builders useful parameters for assessing how it may fit into long-context or tool-using workflows; they do not, by themselves, establish performance on a particular workload.

Mistral lists the model as released on April 30, 2026. It is part of the Mistral models and appears among the newest models.

Key features

Text, image and file inputs

Inputs include text, image and file, while the output modality is text. Vision support is listed, so image input is part of the model’s stated capabilities. Teams should account for the distinction between accepting multimodal inputs and generating multimodal outputs: the listed output here is text.

Long context and reasoning

The 262,144-token context window is the central capacity figure for teams considering lengthy prompts or documents. Reasoning is listed as supported, as is tool use. Those capabilities put it in the categories of vision language models, reasoning models, models with tool calling and longest-context models.

Pricing

Batch pricing is listed at USD 0.75 per 1M tokens for input and USD 3.75 per 1M tokens for output. Output tokens are priced at a higher rate than input tokens, so teams estimating spend should account for both sides of their expected token usage. The listed rates make the input and output split explicit, but actual cost depends on usage volume and the balance between prompt and generated tokens.

Platforms

The model is identified by the provider-facing ID mistralai/mistral-medium-3-5:batch, with Mistral as maker and lab. Its listed modality is text, with text, image and file inputs and text outputs. The batch designation is part of the model entry; the available facts do not specify additional platform or deployment details.

Who it's for

This model is a candidate for teams evaluating instruction-following systems for agentic workflows, coding or complex tasks, particularly where a large context window and tool support matter. Image and file inputs may also be relevant when a workflow needs to pass non-text material into a text-producing model. The specifications are useful for screening fit, but teams will need workload-specific evaluation to decide whether it meets their quality, latency or integration needs.

For teams comparing within Mistral’s lineup, related entries include Mistral Medium 3.5, Mistral Small 4 and Mistral Small 4 (batch).

Pros and cons

Pros

• A 262,144-token context window provides substantial prompt capacity.
• Text, image and file inputs are listed, with vision support and text output.
• Reasoning and tool support align with the model’s stated focus on agentic and complex tasks.
• Input and output token rates are stated separately for straightforward cost estimation.

Cons

• Output pricing is higher than input pricing, so output-heavy usage may warrant careful budgeting.
• The listed output is text, not image or file generation.
• The available specifications do not state performance results, latency or other deployment characteristics.

Alternatives

For coding-oriented comparison, Devstral 2 2512 is another Mistral model to consider. For smaller named model options, the lineup includes Ministral 3 14B 2512, Ministral 3 3B 2512, Ministral 3 8B 2512 and Ministral 3 8B 2512 (batch). The available facts do not provide comparative pricing or performance for these alternatives, so selection should depend on each team’s own requirements and evaluation.

Verdict

Mistral Medium 3.5 (batch) has a clear specification profile: dense 128B instruction following, a 262,144-token context, reasoning and tool support, multimodal inputs, text output and separately stated batch token rates. That makes it worth evaluating for teams whose workflows need those capabilities. The facts establish scope and pricing, not quality on any specific task, so a fit decision should rest on testing against the team’s prompts, tools and cost expectations.