Kimi-Dev-72B
Moonshot's open SWE agent — large-scale RL on real repos in Docker, 60.4% on SWE-bench Verified at launch.
About
Kimi-Dev-72B is Moonshot's purpose-built software engineering model, and its training is unusually honest: the model patches real repositories inside Docker containers and only gets rewarded when the entire test suite passes. No partial credit for plausible-looking diffs. That regime, applied at scale on top of a Qwen2.5-72B base mid-trained on 150B tokens of GitHub issues and PR commits, produced the strongest open-source SWE result of its generation.
At release in June 2025 it hit 60.4% on SWE-bench Verified, the best open-source score at the time. It's MIT-licensed on Hugging Face with a 131K context window, which made it an instant default for local coding agents. Later Moonshot releases have pushed past it, but Kimi-Dev remains the clean permissive baseline for agentic software engineering.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Vendor-reported (HF model card): 60.4% on SWE-bench Verified — SOTA among open-source models at release (June 2025)
- Based on Qwen2.5-72B, mid-trained on ~150B tokens of GitHub issues and PR commits
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: Moonshot AI release (June 16 2025) with SWE-bench Verified 60.4% on the HF model card; MIT license confirmed via HF API; open GitHub repo (MoonshotAI/Kimi-Dev); widely benchmarked by the community. Strengths: permissive license, honest RL training regime, strong adoption in local coding agents. Docked for: (1) headline numbers are vendor-reported; (2) superseded by later Moonshot releases. License: MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Moonshot AI; released 2025-06-16
- · 60.4% SWE-bench Verified (SOTA open at release)
- · Benchmarks are vendor-reported; superseded by later Moonshot releases