Mistral AI's Medium 3.5, a 128-billion-parameter multimodal model, has been available since April 2026 but remains underutilized by Australian developers. Priced at $1.50 per million input tokens, it combines open weights, a 256,000-token context window, and adjustable reasoning capabilities.
TL;DR
- Mistral Medium 3.5 is a 128B-parameter multimodal model with open weights and a large context window.
- It is priced at $1.50 per million input tokens and $7.50 per million output tokens.
- The model is particularly relevant for Australian developers due to data residency and licensing benefits.
What happened
Mistral Medium 3.5 was released on April 28, 2026, as Mistral AI's first 'merged flagship' model, integrating chat, reasoning, tool use, and coding into a single set of weights. It is a dense 128B-parameter architecture that accepts both text and images, with a context window of 256,000 tokens.
The model is priced at $1.50 per million input tokens and $7.50 per million output tokens through Mistral's platform, La Plateforme. It offers a 50% batch-processing discount for non-real-time workloads.
Mistral Medium 3.5 has been relatively quiet in the Australian developer community, despite its capabilities and open weights.
Why it matters
For Australian developers, Mistral Medium 3.5 offers the ability to run the model on local infrastructure, addressing data residency concerns. This is particularly important for projects in health, finance, and government-adjacent sectors.
The pricing is published in USD, which means Australian developers pay the same rate as others but must account for exchange rate fluctuations. The open weights and Modified MIT licence also provide flexibility for commercial use and self-hosting.
The model's adjustable reasoning feature allows it to scale its reasoning effort based on the task, which can help manage latency and token spend more efficiently.
Key facts
- Mistral Medium 3.5 was released on April 28, 2026.
- It is a 128-billion-parameter dense model.
- The model accepts both text and images in the same request.
- It has a context window of 256,000 tokens.
- Pricing is $1.50 per million input tokens and $7.50 per million output tokens.
- A 50% batch-processing discount is available for non-real-time workloads.
- The model is licensed under a Modified MIT licence.
- It achieved approximately 77.6% on the SWE-Bench coding benchmark.
Context
Mistral Medium 3.5 is part of a growing trend of open-weight models that offer flexibility and cost savings for developers. Its multimodal capabilities and large context window make it a strong contender in the AI market.
The model's adjustable reasoning feature is a notable advancement, allowing it to optimize performance based on the complexity of the task. This can be particularly beneficial for developers working on a variety of projects with different requirements.
For Australian developers, the ability to self-host the model on local infrastructure is a significant advantage, addressing data residency and compliance concerns. The open weights and permissive licensing also provide opportunities for customization and commercial use.
