Repository-scale coding agents
Use it when an agent needs to inspect a codebase, call tools, plan changes, and edit multiple files.
Independent tool overview
Devstral at a glance
Devstral was Mistral AI's family of models for software-engineering agents. Mistral deprecated Devstral 2 in May 2026 and recommends Mistral Medium 3.5 for new integrations.
Visit the official Devstral site ↗
- Current replacement
- Mistral Medium 3.5
- Best for
- Coding agents and repository-scale tasks
- Context window
- 256K tokens
- Current API price
- $1.50 input / $7.50 output per 1M tokens
- License
- Modified MIT for current replacement
Overview
What Devstral is
Mistral introduced Devstral in 2025 as a model built for code agents that inspect repositories, use tools, edit multiple files, and work through real software-engineering tasks. The family progressed from the Apache 2.0 Devstral Small releases to Devstral 2 and Devstral Small 2.
The standalone Devstral models are now deprecated. Mistral's current replacement is Mistral Medium 3.5, a 128B open-weight multimodal model optimized for agentic and coding workloads with a 256K context window. It also powers Mistral's current Vibe remote coding agents.
That transition matters for buyers and developers: Devstral remains useful historical context, but new API integrations should target Mistral Medium 3.5 instead of an older Devstral model ID. Teams can use the hosted API or evaluate the released weights under Mistral's Modified MIT license.
Use cases
Who Devstral is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Long-running software tasks
The 256K context window and agent-oriented training suit debugging, refactoring, and multi-step implementation work.
Teams evaluating open weights
Mistral publishes weights for Medium 3.5, giving qualified teams a path beyond a hosted-only API.
Capabilities
Core Devstral features
Agentic coding
Designed for software-engineering workflows that combine reasoning, repository exploration, tool use, and code changes.
256K context
The current replacement can process large code and documentation contexts in one interaction.
Tool and function calling
Supports function calling, structured outputs, agents, conversations, and chat completions through Mistral's APIs.
Multimodal input
Mistral Medium 3.5 expands beyond the original text-focused Devstral line with multimodal capabilities.
Hosted or open-weight deployment
Use Mistral's managed API or evaluate the released weights for controlled deployment requirements.
Process
How the Devstral workflow works
Step 1
Choose the current model
Start new integrations with `mistral-medium-3-5`, not a deprecated Devstral model ID.
Step 2
Connect an agent scaffold
Provide the model with repository access and narrowly scoped tools for search, editing, testing, and version control.
Step 3
Set review boundaries
Require human approval for sensitive commands, dependency changes, deployments, and destructive operations.
Step 4
Evaluate on your codebase
Measure task completion, test pass rate, latency, and token cost on representative engineering work before rollout.
Cost
Devstral pricing and free plan
Mistral Medium 3.5 is usage-priced through the Mistral API. Self-hosting the open weights adds infrastructure and operations costs.
API input
$1.50 per 1M tokens
Hosted input-token pricing for Mistral Medium 3.5.
- Usage based
- 256K context
- API access
API output
$7.50 per 1M tokens
Hosted output-token pricing for Mistral Medium 3.5.
- Usage based
- Agent and chat endpoints
- Batching supported
Open weights
Infrastructure costs apply
Weights are available under Mistral's Modified MIT license.
- Self-managed deployment
- Requires substantial compute
- Review license terms
Pricing checked . Check current pricing at the source ↗
Assessment
Devstral strengths and limitations
Where it stands out
- Purpose-built lineage for tool-using software-engineering agents
- Large 256K context window for repository and documentation context
- Current replacement supports hosted API access and released weights
- Structured outputs and function calling fit production agent workflows
What to consider
- Devstral 2 and Devstral Small 2 are deprecated, so older tutorials and model IDs can be misleading
- Self-hosting a 128B model requires substantial hardware and operational expertise
- Vendor benchmark results do not guarantee performance on a team's own repositories
- Agentic coding still needs permissions, tests, review, and safeguards around tool execution
Compare
Devstral alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
Claude Code
A polished coding agent with strong terminal and repository workflows.
Explore Claude Code →Coding
Cursor
A full AI code editor for developers who want an integrated desktop experience.
Explore Cursor →Agents
Mistral AI
Explore Mistral's broader current model and platform lineup.
Explore Mistral AI →Questions
Devstral FAQs
Is Devstral still available?
The original Devstral line is no longer the recommended choice for new integrations. Mistral marks Devstral 2 as deprecated and directs developers to Mistral Medium 3.5.
What replaced Devstral?
Mistral Medium 3.5 is the official replacement. It is a 128B open-weight multimodal model optimized for agentic and coding work.
How much does the current Devstral successor cost?
Mistral lists Medium 3.5 at $1.50 per million input tokens and $7.50 per million output tokens through its hosted API.
Can I self-host it?
Mistral publishes Medium 3.5 weights under a Modified MIT license. Self-hosting is possible for teams with suitable compute, security, and model-operations capacity.
Is it safe to let a coding agent change production code?
Treat model output as untrusted until tested and reviewed. Limit tool permissions, isolate execution, protect secrets, and require human approval for deployments or destructive actions.
Bottom line
Our Devstral verdict
Devstral established Mistral's coding-agent line, but new users should evaluate Mistral Medium 3.5 instead. It is compelling for teams that want strong agentic coding capabilities plus an open-weight option, provided they can manage the review and infrastructure requirements.
Visit Devstral website ↗Similar Coding Tools
View all Coding AI tools →
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.