Long-running repository tasks
Handle codebase work that benefits from sustained investigation, tool use, testing, and iteration.
Independent tool overview
SWE-1.7 at a glance
SWE-1.7 is Cognition's agentic coding model for long-running software-engineering tasks, available through Devin on web, desktop, and the command line.
Visit the official SWE-1.7 site ↗
- Product type
- Agentic software-engineering model
- Access
- Devin Web, Desktop, and CLI
- Serving speed
- Up to 1,000 tokens per second via Cerebras
- Model base
- Kimi K2.7
- Last reviewed
- August 28, 2026
Overview
What SWE-1.7 is
SWE-1.7 is Cognition's most capable in-house software-engineering model as of this review. It was trained from a Kimi K2.7 base and further post-trained with reinforcement learning for codebase exploration, tool use, and longer-horizon agent tasks.
The model is delivered inside Devin rather than as a general chat model or a standalone public API. Cognition serves it through Cerebras at a stated 1,000 tokens per second across Devin Web, Desktop, and CLI.
Cognition reports near-frontier results on several coding benchmarks at a lower cost, but those are vendor results and some evaluations use different agent harnesses for different model providers. Teams should judge SWE-1.7 on representative repositories, task success, review load, and total spend—not benchmark position alone.
Use cases
Who SWE-1.7 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Cost-conscious Devin users
Choose Cognition's in-house model when the desired balance is strong coding capability with lower platform cost.
Backlog and maintenance work
Apply the model to scoped bug fixes, migrations, refactors, and technical-debt slices with objective acceptance checks.
Teams valuing fast feedback
Use the high-throughput serving path for interactive planning and quicker coding-agent turns.
Capabilities
Core SWE-1.7 features
Long-horizon optimization
Cognition trained SWE-1.7 for asynchronous software tasks that require many turns, tools, and intermediate decisions.
Self-compaction
The model learned to summarize its working state and continue from that summary when a run approaches its context limit.
Codebase exploration
Cognition reports that SWE-1.7 reads and searches more extensively before editing, especially on root-cause bug investigations.
Fast Cerebras inference
The launch announcement states that SWE-1.7 is served at 1,000 tokens per second.
Multiple Devin surfaces
Use the model through Devin's browser product, desktop application, or command-line interface.
Full Devin tool environment
Within Devin, the model can work with an isolated machine, shell, browser, editor, repositories, pull requests, and configured integrations.
Process
How the SWE-1.7 workflow works
Step 1
Connect and prepare the repository
Give Devin repository access, index the relevant branches, and configure dependencies, secrets, and validation commands.
Step 2
Define a verifiable task
Provide a narrow outcome, relevant context, constraints, before-and-after examples, and the tests or checks that determine success.
Step 3
Select SWE-1.7 in Devin
Use the model where it is available in Devin Web, Desktop, or CLI, then review the initial codebase assessment and plan.
Step 4
Monitor the run
Course-correct when requirements are ambiguous and watch for scope expansion, unnecessary refactors, or expensive changes in direction.
Step 5
Review before merging
Inspect the diff, run the full relevant test suite, check security and migration risks, and require human approval for production code.
Cost
SWE-1.7 pricing and free plan
SWE-1.7 is included as a model inside Devin rather than sold with standalone token pricing. Devin offers Free, $20 Pro, $200 Max, and team plans; actual task cost depends on plan quota and on-demand usage.
Free
$0
Try Devin with limited usage.
- One member
- Limited Devin usage
- Includes Devin Review and DeepWiki access
Pro
$20/month
An individual plan for regular use.
- One member
- Daily and weekly quota shared across Devin sessions, Terminal, and Windsurf
- Pay-as-you-go on-demand credits after quota
- Slack, Linear, and MCP integrations
Max
$200/month
A larger individual plan for power users.
- One member
- Larger weekly quota
- No daily cap
- On-demand usage available beyond quota
Teams
From $80/month
A shared plan with unlimited members and a monthly minimum.
- $40 per full seat with a Pro-equivalent quota
- Free flex seats draw entirely from shared on-demand credits
- Full seats include Windsurf access
- On-demand credits are shared across the team
Pricing checked . Check current pricing at the source ↗
Assessment
SWE-1.7 strengths and limitations
Where it stands out
- Cognition reports competitive results across FrontierCode 1.1, Terminal-Bench 2.1, and SWE-bench Multilingual.
- High-throughput serving can make an agent feel substantially more responsive.
- Training explicitly targets long-horizon work, self-compaction, and detailed repository exploration.
- Runs inside the broader Devin environment with repositories, tools, integrations, and pull-request workflows.
- Offers a lower-cost option within Devin than relying only on the most expensive frontier models.
What to consider
- SWE-1.7 is tied to Devin and does not have separately published API token pricing or downloadable weights.
- Benchmark results come from Cognition, and the evaluation harness differs across some compared providers.
- Cognition notes that deeper reasoning can increase change scope, tests, and the number of files touched.
- Public plan pages do not state one fixed number of SWE-1.7 tasks per quota because consumption varies by session.
- Agent-generated changes still require review, testing, security checks, and production ownership by a human team.
Compare
SWE-1.7 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
Claude Code
Consider Claude Code for Anthropic's terminal-centered coding agent and direct access to Claude's current coding models.
Explore Claude Code →Coding
Codex
Consider Codex for OpenAI's multi-surface agentic coding workflow and cloud task execution.
Explore Codex →Coding
Cursor
Consider Cursor when the priority is an AI-native editor with interactive completion and agent workflows.
Explore Cursor →Questions
SWE-1.7 FAQs
What is SWE-1.7?
SWE-1.7 is Cognition's coding model optimized for agentic software-engineering work, especially longer-running tasks that require codebase search, tools, testing, and multiple decisions.
Where can I use SWE-1.7?
Cognition says SWE-1.7 is available in Devin on the web, in Devin Desktop, and through Devin CLI.
How much does SWE-1.7 cost?
There is no standalone SWE-1.7 API price. It is used through Devin plans: Free, Pro at $20 per month, Max at $200 per month, or Teams starting at an $80 monthly minimum, with variable quota and on-demand consumption.
Is SWE-1.7 open source?
No public SWE-1.7 weights or source release is listed. Cognition trained it from the Kimi K2.7 base and serves the resulting model inside Devin.
How does SWE-1.7 perform on coding benchmarks?
Cognition reports 42.3% on FrontierCode 1.1 Main, 81.5% on Terminal-Bench 2.1, and 77.8% on SWE-bench Multilingual. Treat these as vendor-reported results and compare models on your own tasks.
Does SWE-1.7 replace code review?
No. Its changes should be reviewed like any other contribution, with relevant tests, security checks, migration review, and human approval before merge or deployment.
Bottom line
Our SWE-1.7 verdict
SWE-1.7 is a strong option for existing Devin users who want fast, lower-cost agentic coding without giving up much reported benchmark capability. Its real value will depend on task success and review burden in the team's repositories, while users seeking direct model APIs or local deployment should look elsewhere.
Visit SWE-1.7 website ↗Similar Coding Tools
View all Coding AI tools →
ZCode
ZCode - Z.ai's agentic coding environment tuned for GLM-5.2
Grok Build
Grok Build - SpaceXAI's newly open-sourced coding agent and CLI for complex coding work
Cursor for iOS
Cursor for iOS - Cursor's new mobile experience for coding on-the-go
Canva Code
Canva Code 2.0 - Canva's vibe-coding tool with drag-and-drop editing
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.