Cognition

SWE-2

Cognition

SWE-2 is Cognition’s Kimi K3-derived coding model, first publicly introduced and served in Devin Desktop and CLI on September 10, 2026. It extends Cognition’s reinforcement-learning recipe to the roughly 2.8T-parameter scale, optimizing several reasoning-effort levels in one training run. Effort-specific cost penalties target the tradeoff between engineering capability and inference cost, while expanded coding environments and hardened verifiers encourage more reliable task completion. The released identity is the trained model rather than the surrounding Devin application. Web and Fusion rollout were announced separately. Public weights and exact model-serving context/output limits are not established by the launch announcement.

SWE-1.7

Cognition

SWE-1.7 is Cognition’s Kimi K2.7 Code-derived model for long-running software-engineering tasks, released through Devin Web, Desktop, and CLI. Extensive reinforcement learning targets thorough codebase exploration, faithful task execution, and concise reasoning while preserving the ability to think longer on hard problems. Learned self-compaction summarizes working state and resumes from those summaries, allowing agent sessions to continue beyond one model context window without implying an unlimited single-call context. Training uses hardened executable environments and alternating cost penalties. The publisher explicitly distinguishes its behavior from the base checkpoint. Cognition does not publish weights or verified standalone model limits for this release.

SWE-check

Cognition

SWE-check is a specialized bug-detection language model developed by Cognition with Applied Compute and first made available in the Windsurf Next public preview. It analyzes a change diff, investigates the surrounding code, and produces structured bug descriptions and fixes for display in the editor. Reinforcement learning first builds task capability, then tunes latency and product behavior using a separate alignment phase. This is a distinct trained generative review model rather than a generic code-review feature or binary decision endpoint. Architecture, parameter counts, and model context limits remain undisclosed. Its reported parity with frontier models applies to internal in-distribution evaluations, with a remaining gap on out-of-distribution tasks.

SWE-1.6

Cognition

SWE-1.6 is Cognition’s next coding model, first served as a limited-user preview on March 1 and generally released in Windsurf on April 7, 2026. It uses the same underlying pretrained model as SWE-1.5, but the production release was post-trained anew to improve both coding capability and interaction behavior. Training discourages overthinking, repeated reasoning loops, and unnecessary terminal use, while encouraging parallel tool calls. The earlier preview and production checkpoint are grouped under this identity, with their behavioral differences made explicit. Standard and faster serving options provide the same model intelligence; neither establishes publicly downloadable weights or a standalone model API.

SWE-1.5

Cognition

SWE-1.5 is Cognition’s large software-engineering language model, introduced for interactive agent work in Windsurf. Its exact architecture and size are undisclosed, although the publisher describes hundreds of billions of parameters. Model training, inference, and the agent harness were developed together, with reinforcement learning in realistic coding environments and speculative decoding through Cerebras for a faster serving tier. The Falcon Alpha beta identity is a precursor alias, rather than a separate catalog model. It supports repository exploration, application development, and configuration changes through the product’s external tools. Provider speed figures describe its serving stack and do not guarantee end-to-end task latency.

SWE-grep

Cognition

SWE-grep and SWE-grep-mini are Cognition’s language-model specialization for fast multi-turn code-context retrieval. They generate tool calls to grep, glob, and read files, then select relevant files and line ranges for a downstream coding model. Reinforcement learning rewards retrieval quality while encouraging parallel exploration within a small number of serial turns. These are generative agentic search models, distinct from embedding-based retrieval, and were publicly served through Windsurf’s Fast Context subagent. The mini checkpoint is grouped with the same release. Their architecture, parameter counts, and standalone inference limits were not disclosed, and no public weights were announced.

SWE-1

Cognition

SWE-1 is Windsurf’s first model family for the full software-engineering workflow, developed before Windsurf joined Cognition. The main SWE-1 tier is trained for tool-call reasoning and multi-step work in Cascade; SWE-1-lite is a smaller assistant tier replacing Cascade Base, and SWE-1-mini powers rapid predictive code completion in Windsurf Tab. Training emphasizes flow awareness: understanding the shared timeline of developer actions and unfinished work rather than merely generating isolated code snippets. The three tiers are grouped as the original family release. Architecture and parameter counts were not publicly disclosed, and access was through Windsurf rather than downloadable weights.