Darwin-180B-RSI-R3: VIDRAFT's self-improved 180B MoE tops open-weight math reasoning
A 180B-parameter MoE from Korean startup VIDRAFT, trained by recursive self-improvement on its own verified solutions, scoring 73.68 on SuperCLUE's November math and science reasoning board — first among open-weight models.
Why it matters
Only 3B parameters active per token, yet it beats Gemini, Grok and DeepSeek on SuperCLUE math/science; a 4-bit GGUF runs on a 32GB-RAM laptop with no GPU.
What you could build with it
A university research group could run the 180B MoE on a single high-RAM workstation to grade open-ended math and science problem sets, using its strong reasoning scores to pre-score student work before human review.
Does it hold up?
Real but narrow: the SuperCLUE math/science score is independently administered, and the 4-bit GGUF makes it runnable on a laptop — but instruction following (20.69) and long-context (30.32) scores lag well behind closed rivals, so it fits math/science work, not general assistance.
Built with Darwin-180B-RSI-R3
- Darwin-180B-RSI-R3 benchmarks and API guidearticle · Independent benchmark breakdown and API guide, including the MMLU-Pro 88.12 score and the model's instruction-following limits.
- Open-weight MoE scores #2 on SuperCLUE, beating Gemini and Grokarticle · Second writeup covering the SuperCLUE result and the 4-bit POCKET-Darwin GGUF for CPU-only inference.
Learn more
First spotted on article: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.