I wanted to run difficult agentic benchmarks with MiMo v2.6 Distill, like DeepSWE and…
This is a AI post classified by Jev as Open & local models (a model release), kept by the AI Radar because it carries real work, not commentary.
I wanted to run difficult agentic benchmarks with MiMo v2.6 Distill, like DeepSWE and Terminal Bench 4.0. Seems like it will take a week. Much longer than my Qwen3.8 27B runs using the same GPU. Meaning the model does many more turns and/or generate more tokens per turn. Not a big surprise since this inefficiency is one of the symptoms of a model not good enough for this type of tasks. But I'll keep the experiment running.
Posted by Benjamin Marie (6.8k followers) 1 h ago · 7 likes · 490 views · view the original post on X. Kept by the AI Radar as Open & local models.
More AI work like this
- Runs the 2.78-trillion-parameter Kimi K3 model on a single CPU using 8 GB of RAM without… — @tom_doerr
- open weight models just crossed another line — @EHuanglu
- Xiaomi just dropped Mimo 2.6 Pro which claims to be the best open source model. I played… — @deedydas
- So you like Bonsai, — @Mr_WestCoast
- “Hot dog or not?” at 0.4 sec 🌭 — @yoheinakajima
- Mimo V2.6 Flash TP2 Running with a 1.8M KV Pool. — @Tech2Wild
- Meet empero-ai/Qwen3.8-35B-A3B-Distill: a 35B MoE model distilled for reasoning and… — @HuggingModels
- What’s going on here lol what’s the point of these ai reply bots — @lu__jasper
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 41.1k posts from 5k X accounts over the last 14 days, 1.7k tools, 19 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-22 08:01 UTC. Full method.