{"slug": "darwin-180b-rsi-vidraft-s-recursive-self-improvement-moe-model-claims-five-tops", "title": "Darwin-180B-RSI: VIDRAFT's Recursive Self-Improvement MoE Model Claims Five Leaderboard Tops", "summary": "Korean Pre-AGI startup VIDRAFT released Darwin-180B-RSI, a 180-billion-parameter mixture-of-experts reasoning model built on Qwen3.8-Flash-Next that combines selective model merging with recursive self-improvement. The open model, available on Hugging Face, reportedly tops five Hugging Face leaderboards, including a perfect 100% on AIME 2026 and HMMT 2026, 94.44% on GPQA Diamond, 88.12% on MMLU-Pro, and 79.48% on MMMU-Pro. The scores are self-reported with no independent third-party validation, and the model routes only 10 of 512 experts per token, keeping active-parameter compute well below a dense 180B model.", "body_md": "**TL;DR:** VIDRAFT, a Korean Pre-AGI AI startup, has released Darwin-180B-RSI — a 180-billion-parameter mixture-of-experts reasoning model built on Qwen3.8-Flash-Next that combines selective model merging with recursive self-improvement. The model reportedly tops five Hugging Face leaderboards across math, science, general, and multimodal reasoning benchmarks. If the reported gains hold on task-specific workloads, it's worth a closer look for reasoning-heavy applications.\n\nDarwin-180B-RSI is VIDRAFT's open reasoning model, released in late September 2026. Here are the key technical facts as reported:\n\nThe sparse expert activation design (10 of 512 experts per token routing) means that despite the 180B parameter count, the effective compute per forward pass is substantially lower than a dense 180B model — a meaningful practical consideration for deployment cost and latency.\n\nVIDRAFT describes Darwin-180B-RSI as combining two high-level techniques: **selective model merging** and **recursive self-improvement (RSI)**.\n\nAt a conceptual level, the pipeline works roughly like this:\n\nThis approach is conceptually related to ideas like rejection sampling fine-tuning and self-play reinforcement learning, but VIDRAFT's specific implementation details — including iteration count, reward modeling, and filtering criteria — are not disclosed in the source reporting.\n\nThe MoE routing means that at inference time, each token activates only a small subset of the expert network. This is a well-established technique (see Mixtral, DeepSeek-MoE) for scaling parameter count while keeping per-token FLOPs tractable.\n\nVIDRAFT reports the following scores on publicly recognized benchmarks, with Darwin-180B-RSI claiming **first place on five Hugging Face leaderboards**:\n\n| Benchmark | Reported Score | Domain | \n|---|---|---|\n| AIME 2026 | **100% (perfect)** | Mathematical reasoning | \n| HMMT 2026 | **100% (perfect)** | Mathematical competition | \n| GPQA Diamond | **94.44%** | Graduate-level science Q&A | \n| MMLU-Pro | **88.12%** | Multi-domain professional knowledge | \n| MMMU-Pro | **79.48%** | Multimodal understanding & reasoning | \n\n⚠️ **Important caveat:** These numbers are self-reported by VIDRAFT. The source article explicitly notes that no independent third-party validation of these results has been performed. Leaderboard rankings reflect performance on fixed benchmark datasets and may not generalize to production workloads, novel problem distributions, or domain-specific tasks. Engineers evaluating this model should run it against their own held-out test sets before drawing conclusions about real-world reasoning quality.\n\nThe source reports Darwin-180B-RSI is an **open model available on Hugging Face**. If VIDRAFT follows its prior release pattern (the company had a prior Hugging Face feature covered in September), the model weights should be downloadable via the standard Hugging Face toolchain.\n\nTo check availability and download weights once the repository is public:\n\n```\n# Search for the model on Hugging Face\nhuggingface-cli search vidraft/darwin-180b-rsi\n\n# Download when available\nhuggingface-cli download vidraft/darwin-180b-rsi\n```\n\n**Note:** The exact repository path is not confirmed in the source. Search for `VIDRAFT` or `Darwin-180B-RSI` directly on [huggingface.co](https://huggingface.co) to find the correct model card and any associated inference instructions.\n\nNo GitHub repository, API endpoint, or OpenAI-compatible serving URL is mentioned in the source reporting. Check VIDRAFT's Hugging Face profile and official channels for inference API access details.\n\n**Q: Is Darwin-180B-RSI practical to self-host given its 180B parameter count?**\n\nA: The MoE architecture activates only 10 of 512 experts per request, so active-parameter FLOPs are much lower than a dense 180B model. That said, you still need sufficient memory to load all expert weights. Quantized or offloaded serving strategies (e.g., `llama.cpp`, `vLLM` with expert offloading) may be necessary depending on your hardware. Check the model card for official serving recommendations once it is published.\n\n**Q: How does the recursive self-improvement approach differ from standard RLHF or SFT?**\n\nA: Standard supervised fine-tuning (SFT) trains on human-labeled data; RLHF uses a human preference reward model. VIDRAFT's RSI loop instead uses the model's own outputs filtered by verifiable correctness — closer in spirit to rejection sampling fine-tuning or outcome-reward RL (like GRPO or STaR), but without requiring a separate reward model or human annotators for each iteration. The practical implication is that it scales more easily to domains with automated answer verification, like math and formal science.\n\n**Q: Should I trust the perfect AIME and HMMT scores at face value?**\n\nA: Treat them as a signal worth investigating, not a guarantee. Perfect scores on competition math benchmarks from self-reported results warrant independent replication. Run the model on problems from your domain and measure it yourself before committing to it for production use cases.\n\n*Originally reported by AI Market Watch (미국) (2026-09-28) — [source article](https://www.ai-market-watch.com/news/vidrafts-recursive-self-improvement-model-tops-five-global-ai-leaderboards-mctpo0).*", "url": "https://wpnews.pro/news/darwin-180b-rsi-vidraft-s-recursive-self-improvement-moe-model-claims-five-tops", "canonical_source": "https://dev.to/ai_openfree_b23025ef075cf/darwin-180b-rsi-vidrafts-recursive-self-improvement-moe-model-claims-five-leaderboard-tops-3ela", "published_at": "2026-09-29 23:01:18+00:00", "updated_at": "2026-09-29 23:16:59.423431+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "ai-startups", "ai-products"], "entities": ["VIDRAFT", "Darwin-180B-RSI", "Qwen3.8-Flash-Next", "Hugging Face", "Mixtral", "DeepSeek-MoE"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/darwin-180b-rsi-vidraft-s-recursive-self-improvement-moe-model-claims-five-tops", "markdown": "https://wpnews.pro/news/darwin-180b-rsi-vidraft-s-recursive-self-improvement-moe-model-claims-five-tops.md", "text": "https://wpnews.pro/news/darwin-180b-rsi-vidraft-s-recursive-self-improvement-moe-model-claims-five-tops.txt", "jsonld": "https://wpnews.pro/news/darwin-180b-rsi-vidraft-s-recursive-self-improvement-moe-model-claims-five-tops.jsonld"}}