SWE-2
- Price
- Freemium
- API
- Yes
- Weights
- No
WHAT IT DOES
SWE-2 is Cognition's flagship coding model, released September 12, 2026 and built by reinforcement-learning post-training on top of Moonshot AI's 2.8-trillion-parameter open-weight Kimi K3. Cognition reports it matches Claude Fable 5.1 on FrontierCode 1.1 Main (both around 50%) at roughly 64% lower inference cost, scores 73.0% on DeepSWE 1.1 and 92.8% on Terminal-Bench 2.1, but trails Fable 5.1 and GPT-6 Astra by around 30 points on the harder Terminal-Bench 4. It ships with no open weights and no self-serve public API: it is exclusive to Devin's desktop app and CLI, with a Devin Web and Fusion rollout planned. Usage inside Devin is free through mid-October 2026; after that it requires a paid Devin plan, and separate enterprise credit-based API access is priced at $3.00 per million input tokens, $0.30 per million cached input tokens, and $15.00 per million output tokens.
WHY WE PICKED IT
SWE-2 is a genuine data point in the trend of labs fine-tuning open-weight Chinese base models to compete with closed frontier coding models on price, and it is the model Cognition is betting Devin's entire value proposition on. The catch is access: every benchmark above is Cognition's own, unverified independently, and there is no way to use the model outside Devin's ecosystem.
QUICK FACTS
STRENGTHS
- ✓Matches Claude Fable 5.1 on FrontierCode 1.1 Main at roughly 64% lower reported cost
- ✓73.0% on DeepSWE 1.1 and 92.8% on Terminal-Bench 2.1
- ✓Built via RL post-training on the open Kimi K3 base rather than trained from scratch
- ✓Free to use inside Devin desktop and CLI through mid-October 2026
- ✓Signals a real price/performance path for fine-tuned open-weight bases
LIMITATIONS
- ⚠No open weights and no self-serve public API, usable only inside Devin
- ⚠Trails Claude Fable 5.1 and GPT-6 Astra by roughly 30 points on Terminal-Bench 4, its clear weak spot
- ⚠All benchmark figures are Cognition's own, with no independent third-party verification yet
- ⚠Free usage window is temporary; ongoing use requires a paid Devin plan or enterprise API credits
GETTING STARTED
Use SWE-2 through Devin's desktop app or CLI; it is selectable as a model option and is free to run there through mid-October 2026. After that, it requires a paid Devin plan (Pro starts at $20/month). It is not available as a standalone self-serve API or as downloadable weights.
OFFICIAL LINKS
SIMILAR TOOLS
PRICING
Free tier includes limited features. Paid plans unlock full access, higher usage limits, and commercial usage rights.
FEATURED IN GUIDES
How to Review an Agent Skill Before Installing It
A practical review workflow for third-party agent skills: inspect instructions, scripts, dependencie...
CLAUDE.md In Practice: What Actually Earns A Place In The File
The most-starred CLAUDE.md on GitHub has four rules, not twelve, and the error rates everyone quotes...
What is AI Coding? Complete Guide 2026
AI coding explained: how AI-powered IDEs and coding assistants work and transform software developme...
How to Use AI Coding Tools: Complete Guide 2026
Using AI coding tools effectively: step-by-step guides for Cursor, Google Antigravity, GitHub Copilo...
AI Coding Tools: Which Ones Actually Help Developers?
Best AI coding tools compared: Cursor, Google Antigravity, GitHub Copilot, Replit, and more. Feature...
EXPLORE ALTERNATIVES
Compare SWE-2 with 5+ similar ides & coding tools AI tools.
FREQUENTLY ASKED QUESTIONS
Is SWE-2 free?
SWE-2 offers a free tier with optional paid upgrades.
Does SWE-2 have an API?
Yes, SWE-2 offers an API for programmatic integration.
What is SWE-2 best for?
SWE-2 is best for Cost-Efficient Coding Inside Devin. SWE-2 is Cognition's flagship coding model, released September 12, 2026 and built by reinforcement-learning post-training on top of Moonshot AI's 2. SWE-2 is a genuine data point in the trend of labs fine-tuning open-weight Chinese base models to compete with closed frontier coding models on price, and it is the model Cognition is betting Devin's entire value proposition on. The catch is access: every benchmark above is Cognition's own, unverified independently, and there is no way to use the model outside Devin's ecosystem.
What platforms does SWE-2 support?
SWE-2 supports local, api.
Is SWE-2 open source?
No, SWE-2 is not open source.
How do I get started with SWE-2?
Use SWE-2 through Devin's desktop app or CLI; it is selectable as a model option and is free to run there through mid-October 2026. After that, it requires a paid Devin plan (Pro starts at $20/month). It is not available as a standalone self-serve API or as downloadable weights.
FEATURED ON CURATEDAI
Work on SWE-2? You're hand-reviewed in our directory. Add this badge to your site — it links back to this profile.
