Devin CLI is the command-line edition of Cognition’s autonomous coding agent, Devin, made to run directly in your terminal. You can switch between frontier models from Anthropic, OpenAI, Google, and Cognition mid-session. Reading, writing, and running code all happen locally on your own machine, but long-running work can be handed off to Devin in the cloud, so processing continues even after you close your laptop. It launched on April 27, 2026, and is designed so that your existing MCP servers, skills, and plugins can be brought over as-is.
Key Features
- Switch between models from multiple vendors: In addition to models from Anthropic, OpenAI, Google, and Cognition, you can also choose open-weight models. Because you can switch with the
/modelcommand without interrupting the session, you can use a cheap model for investigation and a high-performance one for implementation - Fusion (lead + sidekick setup): Pairs a high-performance lead model with a low-cost sidekick model to hold down costs while maintaining accuracy. It was introduced in September 2026
- Handoff to the cloud: The
/handoffcommand passes a session you started locally to an agent on Devin Cloud. You can step away while a long build or test run continues - Commands for fine-grained control of the conversation:
/planto agree on an approach before implementation,/askfor read-only investigation that does not modify code,/loopto automatically review diffs,/forkto branch a conversation and/revertto rewind it, and/resumeto pick up an interrupted session - Reuse your existing extensions: You can bring over MCP servers, skills, hooks, and plugins built for Claude Code. Switching tools does not mean rebuilding your configuration
- Lightweight runtime written in Rust: The core is written in Rust, installs in under a minute, and starts quickly
Pricing
| Plan | Monthly price | What’s included |
|---|---|---|
| Free | $0 | Light quota, limited model availability. Unlimited inline edits and Tab completions |
| Pro | $20 | Full access to frontier models from OpenAI, Claude, and Gemini, plus Devin Cloud |
| Max | $200 | Everything in Pro, with a higher quota |
| Team | $80 + $40 per seat | Sharing and collaboration, consolidated billing, admin dashboard, priority support |
| Enterprise | Contact sales | VPC deployment, SSO via SAML/OIDC, a dedicated account team |
On the Pro plan and above, Cognition’s SWE-2 Free is listed as available at no cost in Devin Desktop and the CLI through October 10, 2026. Since this is a time-limited arrangement, it is worth checking whether the regular quota will be enough before you commit for the long term. Note that pricing is shared across the whole Devin brand (Desktop / CLI / Cloud) and is not set for the CLI alone.
Pricing is as of September 2026. Please check the official site for current pricing.
Pros & Cons
✅ Pros
- You are less at the mercy of whether a given model happens to do well. When something is not working, you can change only the model and retry with the same context instead of throwing away the conversation
- Waiting time becomes your own time again. Long tasks can be handed to the cloud so you can close your laptop rather than sitting in front of the screen until the run finishes
- Configuration you built up in other tools does not go to waste. Because MCP servers, skills, and plugins carry over, the upfront cost of switching is small
- Thanks to Fusion, you can aim for practical accuracy without running the most expensive model every time, which keeps costs from piling up on everyday small requests
/asklets you request read-only investigation, reducing the worry that code you are not ready to touch will be rewritten without warning
⚠️ Cons
- The Free plan limits which models you can use, so Pro or above is effectively a prerequisite for real work
- Handoff to the cloud and use of Devin Cloud are restricted unless you are on a higher plan
- Because it shares branding and pricing with Devin itself (Desktop / Cloud), it is sometimes hard to tell where the CLI’s capabilities end
- Headline perks such as free SWE-2 usage are time-limited, so you should assume the terms may change
Comparison with Similar Services
| Service | Provider | How it differs from Devin CLI |
|---|---|---|
| Claude Code | Anthropic | Fixed to Claude-family models. Devin CLI lets you choose models across multiple vendors |
| Codex CLI | OpenAI | Fixed to GPT-family models. Devin CLI can delegate a session to a cloud-side agent |
| Gemini CLI | Fixed to Gemini models. Devin CLI combines two models with Fusion to hold down costs |
Who Is It For
- Developers who want to compare AI models in practice and pick the best combination for each project
- People who want to work through bug fixes and new features interactively in a local repository
- People who want to hand long-running tasks to an agent and move on to something else in the meantime
- Those who have already set up MCP servers and skills and want to try a different CLI without losing that investment
- Teams who want to run one-off commands with
devin -pin CI or scripts to build an automation pipeline
Summary
Terminal-based coding agents are now available from several vendors, but most of them are locked to their provider’s own models. Devin CLI opens that up, letting you switch between models from multiple vendors within a single conversation. On top of that, it connects local execution and cloud handoff in one continuous session. Start with the Free plan to get a feel for the interface and command set, then consider Pro or above once you need the wider model selection and cloud integration.