Model Releases
With only ~$1,000 of training compute, we trained ๐ณ๐ฟ๐ผ๐บ ๐๐ฐ๐ฟ๐ฎ๐๐ฐ๐ต a 1.7B coding agent that beats Claude 3 Opus and GPT-4o on SWE-benโฆ
With only ~$1,000 of training compute, we trained ๐ณ๐ฟ๐ผ๐บ ๐๐ฐ๐ฟ๐ฎ๐๐ฐ๐ต a 1.7B coding agent that beats Claude 3 Opus and GPT-4o on SWE-bench Verified ๐คฏ At 25.5% pass@1, it sets a new record for nanoswe, our op
With only ~$1,000 of training compute, we trained ๐ณ๐ฟ๐ผ๐บ ๐๐ฐ๐ฟ๐ฎ๐๐ฐ๐ต a 1.7B coding agent that beats Claude 3 Opus and GPT-4o on SWE-bench Verified ๐คฏ At 25.5% pass@1, it sets a new record for nanoswe, our open SWE-bench speedrun. How is this even possible? ๐งต๐
Related
- Kimi K2.6 just dropped. And it crushed Claude Opus 4.6 on SWE-Bench Pro. Kimi K2.6: 58.6 GPT-5.4 xhigh: 57.7 Gemini 3.1 Pro: 54.2 Claude Opuโฆ
- I ran GLM 5.2 with OpenCode harness against Claude Opus this week deployed locally. Bottom line: It is a real frontier coding model and insaโฆ
- dear god lol - new kimi model is a fucking beast. GPT 5.4 level coding, 76% cheaper than opus 4.7 and 100% open source / free to use, i meanโฆ
- We show how you can use the River API to fine-tune open weight models to beat GPT-6 Astra and Claude Opus 5.5 on tasks you care about, with โฆ
Source: Clem Delangue (X) | 2026-10-12