Hey Music Lovers ! You have a track and you want the stems. Retro 6stems splits any audio file into 6 clean, separated tracks Drop an MP3, WAV, FLAC or M4A β Demucs gets to work immediately with a minimal retro design . Instrumental mix included automatically Vocals Drums Bass Guitar Piano Other (strings, synths...) Preview each stem directly in the browser, download individually or grab the full ZIP. No install. No GPU needed on your end. Just upload and wait.
𧬠Darwin-35B-A3B-Opus β The Child That Surpassed Both Parents
What if a merged model could beat both its parents? We proved it can. Darwin-35B-A3B-Opus is a 35B MoE model (3B active) built with our Darwin V5 engine β the first evolution system that CT-scans parent models before merging them. π€ Model: FINAL-Bench/Darwin-35B-A3B-Opus
The result speaks for itself: GPQA Diamond 90.0%, versus Father (Qwen3.5-35B-A3B) at 84.2% and Mother (Claude 4.6 Opus Distilled) at 85.0%. That's +6.9% over Father and +5.9% over Mother. Not a tradeoff β a genuine leap. Meanwhile, MMMLU sits at 85.0% (Father: 85.2%), multimodal is fully intact, and all 201 languages are preserved.
How? Model MRI changed everything. Traditional merging is guesswork. Darwin V4 added evolution. Darwin V5 added X-ray vision. Model MRI scans each parent layer by layer and discovers: Mother's L34βL38 is the reasoning engine (peak cosine distance), 50β65% of Mother's experts are dead (killed by text-only distillation), and Father is a healthy generalist with every expert alive. The prescription: transplant Mother's reasoning brain at L38 (90% weight), replace her dead experts with Father's living ones, and let Father's router handle the output layer. Reasoning went up. Versatility stayed intact. No tradeoff β just evolution.
35B total, 3B active (MoE) Β· GPQA Diamond 90.0% Β· MMMLU 85.0% (201 languages) Β· Multimodal Image & Video Β· 262K native context Β· 147.8 tok/s on H100 Β· Runs on a single RTX 4090 (Q4) Β· Apache 2.0 Darwin V5's full algorithm and technical details will be released alongside an upcoming paper.
I fine-tuned Qwen2.5 with GRPO to actually think before it answers β not just pattern-match.
Most LLMs mimic reasoning. This one builds a real cognitive path:
π Plan β understand the task π Monitor β reason step by step β Evaluate β verify before answering
Every response follows a strict structured protocol: <think> <planning> ... <monitoring> ... <evaluation> ... </think> Then a clean, reasoning-free <output>.
The model self-checks its own structure. If a section is missing or malformed β the response is invalid.
This isn't chain-of-thought slapped on top. The reasoning protocol is baked in via RL.