Cognition Debuts SWE-1.7 Coding Model in Devin

Cognition launched SWE-1.7 on Wednesday, an AI model for software engineering tasks that the company describes as its most capable to date. The model is integrated into Devin — Cognition's autonomous engineering agent — across Web, Desktop and CLI, and is available via Cerebras at 1,000 tokens per second.

Cognition said SWE-1.7 delivers frontier-level intelligence at a lower cost than prior approaches. The model was trained from a Kimi K2.7 base using a globally distributed reinforcement learning (RL) pipeline spanning four data centers across three continents, combining Cognition's own GPUs with compute from inference providers such as Fireworks.

Unpacking SWE-1.7: Feature Breakdown

Key capabilities Cognition claims for SWE-1.7 include:

Capability	Description
Devin integration	Available across Web, Desktop and CLI interfaces
Multi-cluster RL training	Distributed across four datacenters on three continents
Self-compaction	Model summarizes working state to extend task horizons
Data-quality pipeline	Automated tests filter low-signal tasks and prevent reward-hacking

AI Coding Agents & Automated Developer Workflows

AI coding agents now execute multi-step workflows autonomously — from planning through deployment — with minimal human oversight.

Still, the gap between legacy coding tools and modern agentic AI systems is significant. Where earlier tools suggested snippets, today's agents maintain session context, coordinate across codebases and handle end-to-end delivery tasks.

Anthropic, OpenAI and Google are building systems that automate software delivery through multi-agent coordination. These agents generate application scaffolding, write and run test suites, flag integration issues and propose deployment configurations 6 all from plain-English instructions.

Orchestration Over Raw Intelligence

As agents reason across steps and act on external systems, the engineering challenge shifts from model capability to AI orchestration. Governance mechanisms — combining machine learning operations, cybersecurity controls and escalation paths —are essential to prevent runaway automation at scale.

Cognition in the News

Cognition reached a $26 billion post-money valuation after raising more than $1 billion in a May 2026 Series D led by Lux Capital, General Catalyst and 8VC. The round disclosed a $492 million annualized revenue run rate, with enterprise Devin usage growing roughly 50% month-over-month — a trajectory that traces back to Goldman Sachs deploying Devin as its first AI "employee" in July 2025 and Cognition's acquisition of Windsurf that same month.

Enterprise distribution has since locked in through major SI partnerships. Infosys announced a collaboration in January 2026 to embed Devin into its Topaz Fabric AI stack, followed days later by Cognizant integrating Devin and Windsurf with its Flowsource platform.

Cognition also shipped Devin 2.2 in February 2026, adding Computer Use, Self-Verification and closed-loop Autofix capabilities.