LLM Reference

SWE-2

Released
2026-09-10
Last refreshed
2026-09-10
Status
Researched today
ProprietaryCommercial use: conditionalCodingAgentsJSON / Tool useCodingAgents

SWE-2 is worth evaluating for coding, agents, and json / tool use when its provider route and context window match the workload.

Use it for

  • Teams evaluating coding, agents, and json / tool use
  • Buyers comparing 1 tracked provider route

Do not use it for

  • Vision or document-understanding workloads
Specifications
Family
SWE-2
Released
2026-09-10
Parameters
2.8T (base: Kimi K3)
Specialization
coding
Openness
Proprietary
License
ProprietaryCommercial use: conditional
Weights
Not released
Code
Unknown
Training
Reinforcement Learning
Created by

Building autonomous AI software engineers.

San Francisco, California, United States
Founded 2023
Website
Pricing
Output / 1M
-
Input / 1M
-

Cheapest of 1 route · Devin

About

SWE-2 is Cognition's proprietary coding model for long-horizon asynchronous software-engineering tasks in Devin, launched September 10, 2026. Post-trained from the Kimi K3 2.8T-parameter base with additional RL at multi-trillion-parameter scale, SWE-2 pushes the cost–performance Pareto frontier and is available through Devin Desktop and CLI at launch, with rollout to Devin Web and Fusion.

Top use-case fit: coding, agents, and build tasks

Coding

Included by capability and metadata signals in the decision map.

Agents

Included by capability and metadata signals in the decision map.

JSON / Tool use

Included by capability and metadata signals in the decision map.

Provider price ladder

Compare API pricing across 1 providers for input and output tokens, batch, and cached reads when available.

ProviderInput / 1MOutput / 1MRoute
Devin--
ServerlessPartial

Capabilities

ReasoningJSON / Tool useCode Execution

Benchmark peer barsfor Coding

No task-mapped benchmark peers are available for this model yet.

Migration checks

No linked migration route is available for this model yet.