SWE-2: Benchmarks, Pricing & Context Window
SWE-2 is a language model from Cognition, released in September 2026.
SWE-2 is Cognition's most advanced coding model, post-trained from Moonshot AI's 2.8-trillion-parameter Kimi K3 with reinforcement learning scaled to the multi-trillion-parameter regime for the first time. A single RL run trains all
SWE-2 benchmarks
Capability tiers
Standing within each category, adjusted for leaderboard depth.
Real tasks performance
High-confidence performance for SWE-2 across real-world prompt categories. Only 95% intervals at most 4 points wide are shown.
Performance by conversation depth
How SWE-2 holds up as conversations get longer.
Quality Tracker
SWE-2 Performance Across Datasets
Scores sourced from the model's scorecard, paper, or official blog posts
llm-stats.com - Fri Nov 06 2026Try now
huggleMake it with
SWE-2.
SWE-2 latency
SWE-2 time to first token, sustained output throughput, and failed-request rate from live model usage over the trailing 7 days.
SWE-2 examples
Recent arena outputs from SWE-2, picked from the highest-ranked matchups.
SWE-2 license
SWE-2 is a proprietary model available under its provider's product and API terms.
- License
- Proprietary
- Hosted access
Proprietary license - usage restrictions apply
SWE-2 resources
Official sources for SWE-2: official playground, official launch post.
SWE-2 vs other models
The most-compared alternatives to SWE-2 are Claude Opus 4.8, Claude Sonnet 5.5, GPT-5.6 Sol. Open any pair side-by-side for benchmarks, pricing, context, and latency.
Models like SWE-2
Models ranked just above and below SWE-2 by LLM Stats score.
FAQ
Common questions about SWE-2.