Darwin-180B-RSI is a 180-billion-parameter vision-language model built from Qwen3.8-Flash-Next. Its publisher says it kept the parent’s 512 routed experts, router, and vision encoder, while updating selected attention paths and shared experts. The release describes the changed share as 0.02%, but that percentage has not been independently audited in the sources available here.
The notable idea is selective adaptation, not shrinking a 180B model into a small one: preserve most of the parent’s architecture and use a self-training loop based on answers checked against references. The publisher reports benchmark gains, but those figures are not independent reproductions. And although a 4-bit version has been run on a laptop, its 111 GB checkpoint relies on SSD streaming rather than fitting in system memory.
What Darwin-180B-RSI is—and what the 0.02% means
Darwin-180B-RSI is a derivative of Qwen3.8-Flash-Next, which the model card identifies as a 180B mixture-of-experts (MoE) vision-language model. The publisher describes Darwin as changing selected components while retaining the parent’s broad architecture. It characterizes the updated portion as 0.02%; the model card does not provide an independently audited measurement of that fraction, so treat it as the publisher’s claim rather than a verified parameter count. FINAL-Bench / VIDRAFT model card
In an MoE model, the total parameter count is not the same as the number of parameters used for every token. A routing system selects experts for a given input, which can lower per-token computation relative to activating every parameter. That sparsity does not make the stored model weights correspondingly small: the 4-bit checkpoint discussed below still occupies 111 GB.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall#1 Best Overall
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
Which parts changed?
According to the publisher, Darwin retained the parent’s 512 routed experts, router, and vision encoder. It updated selected attention paths and the shared expert instead. That is the basis for the description of a targeted update: the routed experts and the mechanism selecting them were left intact, while specific attention and shared-expert components were adapted.
The 0.02% figure should not be read as an independently confirmed audit, a claim that only 0.02% of the stored file differs, or a statement about active parameters per token. It is the publisher’s characterization of the changed share; those are distinct quantities.
How the recursive self-improvement loop is described
The model card describes a training loop in which the model works through practice problems, checks answers against references or executable tests, and uses only correct reasoning for further training. The improved model can then attempt another round of problems. The publisher says the practice sets were filtered against evaluation sets using an 8-gram overlap check and that no human-written reasoning traces were used. These are descriptions from the model publisher, not independently verified properties of the training process.
Rank #2
- Built for Local AI Development: AMD Ryzen AI Halo is designed for local AI development and inference, featuring 128GB unified memory and support for up to 200B parameter models to build and run intensive AI workloads locally.
- 128GB Unified Memory: Features 128GB LPDDR5x unified memory at 8000 MT/s with 256 GB/s memory bandwidth, providing a shared memory pool across the CPU, GPU, and NPU to support larger AI models.
- AMD Ryzen AI Max+ 395 Processor: Features 16 cores, 32 threads, and Zen 5 architecture, paired with AMD Radeon 8060S integrated graphics featuring 40 RDNA 3.5 compute units and an AMD XDNA 2 NPU with up to 50 TOPS.
- Linux AI Developer Platform: Purpose-built for Linux-based AI development with full AMD ROCm software support and preloaded tools, models, and workflows optimized for local AI development.
- Compact, Connected Design: Includes a 2TB M.2 SSD, 10GbE LAN, Wi-Fi 7, Bluetooth 5.4, USB-C connectivity, and HDMI 2.1b.
- Generate solutions: the model attempts practice problems described as previously unseen.
- Verify answers: answers are checked against references or executable checks.
- Train on checked reasoning: the publisher says only reasoning associated with correct answers is used for further training.
- Repeat: a resulting model can be used for another round of problem-solving and training.
Here, “recursive self-improvement” refers to model-level training using the model’s generated solutions after verification. It is not the same as a prompt or tool harness that evolves its instructions, and it is not the training-free model-merging method described in a separate Darwin Family paper. That paper presents an evolutionary merging framework, including an adaptive merge genome and cross-architecture mapping, with reported experiments at 4B–35B scales; it is not evidence that the 180B RSI model was produced by the same training-free procedure. Darwin Family paper
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What the publisher reports on benchmarks
The figures below are reported by FINAL-Bench / VIDRAFT in the Darwin-180B-RSI model card. They should not be treated as independent replications, and the card cautions that comparison settings can differ.
| Measure | Darwin-180B-RSI | Parent model | Attribution and qualification |
|---|---|---|---|
| GPQA Diamond | 94.44% | Not stated in the cited Darwin-180B-RSI card | Publisher-reported score; comparison settings may differ. |
| MMLU-Pro accuracy | 88.12% | 88.04% | Publisher-reported comparison with Qwen3.8-Flash-Next. |
| Mean reasoning length on MMLU-Pro | 3,833 tokens | 4,320 tokens | Publisher-reported figures; the card characterizes the reduction as 11%. |
The MMLU-Pro accuracy difference is small in the reported comparison, while the reasoning-length figures suggest fewer tokens on average for Darwin in that evaluation. Neither result alone establishes a broad improvement across tasks or identical evaluation conditions.
Rank #3
- EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
R3 is a separate follow-up checkpoint
Darwin-180B-RSI-R3 is a later version, and its results should not be substituted for the original Darwin-180B-RSI figures. Its model card says R3 was trained from R1 using 714 correct solutions drawn from 462 boundary problems. On a held-out set of 1,000 SuperGPQA questions, it reports a +1.03-point paired mean-of-four difference from R1, with a 95% confidence interval of [+0.05, +2.00]. The card says GPQA differences were within noise. Darwin-180B-RSI-R3 model card
These R3 results describe a particular follow-up comparison and evaluation protocol; they do not independently validate the initial release’s 0.02% claim or establish that every benchmark improves.
Can you run Darwin-180B locally?
Yes, a local run has been demonstrated on personal hardware, but it is a storage-streaming setup rather than a model that fits wholly into laptop RAM. A Hugging Face Blog article published October 4, 2026 reports a 111 GB 4-bit GGUF checkpoint run on a laptop with 32 GB of RAM and an 8 GB GPU. With SSD streaming, that configuration generated at 4.17 tokens per second. The article recommends at least 120 GB of free NVMe space and notes that storage, context size, prompt processing, and workload affect the experience. Hugging Face Blog: What It Really Takes to Run a 180B MoE Model on Personal Hardware
The practical distinction is between having enough memory to keep weights resident and streaming weights from storage as needed. The reported laptop configuration cannot hold the full checkpoint in RAM. A separate in-memory test in the same article used a 16-thread AMD EPYC CPU setup, reporting 18.4–21.0 tokens per second and 78.8 GB peak memory; it is not directly comparable to the SSD-streamed laptop result.
- Storage: plan for the 111 GB checkpoint and the article’s recommendation of at least 120 GB free NVMe space.
- Memory arrangement: expect SSD streaming on the cited 32 GB RAM laptop configuration, rather than all weights resident in RAM.
- Responsiveness: the reported 4.17 tokens per second is specific to the stated laptop setup; long reasoning outputs can take time, and other workloads or settings may differ.
License and what the evidence establishes
The Darwin-180B-RSI and R3 model cards identify the Qwen Community License 1.0, inherited from the parent model. Review the license itself for the terms applicable to your planned use; the weights should not be assumed to be unrestricted public-domain material. Darwin-180B-RSI model card · R3 model card
The strongest supported conclusion is that the publisher describes a targeted adaptation of a 180B parent and reports a verification-based self-training process and benchmark results. The exact 0.02% share and performance claims remain publisher-reported in the cited sources; the deployment article establishes that one quantized configuration can run with SSD streaming on specified hardware, not that local use is effortless or fast in every setup.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
