Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems
Darwin-180B-RSI
Open 180 billion parameter reasoning model from VIDRAFT that reads text and images and gives a confidence score
About Darwin-180B-RSI
Reasoning with a built-in confidence check is the pitch of Darwin-180B-RSI, an open-weight model from VIDRAFT, a Korean AI company. It went onto Hugging Face on 28 September 2026, and ZDNet Korea and AI Times wrote about it that day. It accepts text and images.
A fine-tune of Qwen3.8-Flash-Next
Darwin starts from Alibaba's Qwen3.8-Flash-Next, a mixture of experts with 180 billion parameters in which 10 of 512 experts work on each token, and it retains the parent's vision encoder and its window of about 262,000 tokens. The model card gives 180 billion parameters, though ZDNet Korea describes it as a 397-billion-class model, so reports differ on its size.
Recursive self-improvement
VIDRAFT's name for its method is recursive self-improvement. The model works through practice problems, answers are kept only when they match an answer key, and the model is then retrained on the reasoning behind them. In VIDRAFT's own comparison with the parent model, accuracy barely changes (88.12 against 88.04 per cent on MMLU-Pro) but average reasoning length falls by about 11 per cent, which makes the model cheaper to run.
Confidence before the answer
A readout VIDRAFT calls zero-token confidence estimates how likely an answer is to be right before the model produces it. An application can use that to escalate a hard case or decline to answer.
Reported results, with caveats
VIDRAFT cites 100 on both AIME 2026 and HMMT February 2026, plus 94.44 on GPQA Diamond and 88.12 on MMLU-Pro, with 79.48 on MMMU-Pro, and describes the set as first place across five Hugging Face leaderboards. The entries are submitted by the publishers themselves, most figures are majority votes over as many as 16 attempts with a large thinking budget, and other models' entries use different settings, so the comparison is not like for like. An independent evaluation had not appeared at the time of checking.
Running it
Commercial users should read the Qwen Community License 1.0 that the weights carry. Self-hosting with vLLM is what the model card describes, and it names no hosted API.
TrustList summary
VIDRAFT, a Korean AI company, put Darwin-180B-RSI on Hugging Face on 28 September 2026 as an open-weight reasoning model. It is a fine-tune of Alibaba's Qwen3.8-Flash-Next and reads both text and images. Its results are the developer's own, and no independent evaluation had been published when the sources were read.
Features
- Reasoning on text and images: keeps the parent model's vision encoder and a context of about 262,000 tokens.
- Self-checked training: the model solves practice problems and keeps only answers that match an answer key, then retrains on them.
- Confidence readout: a score given before the answer estimates the chance it is correct, letting an application escalate or decline.
- Shorter reasoning: the developer reports reasoning about 11 per cent shorter than its parent, which reduces the work needed per answer.
- Self-hosting: the model card describes serving it with vLLM and names no hosted API.
Who it serves
Teams that can host a 180-billion-parameter model themselves and want a reasoning model that reports its confidence before it answers.
Key facts
- Developer: VIDRAFT, a Korean AI company.
- Size: 180 billion parameters on the model card, although press reports disagree on the figure.
- Parent model: Alibaba's Qwen3.8-Flash-Next.
- Licence: Qwen Community License 1.0, which a business should read before commercial use.
Summary written by TrustList from the business's own published information, Nov 2026.
Benchmarks & AI stats
| Benchmark | Official | Community avg |
|---|---|---|
| No benchmark scores yet. Be the first to add one. | ||
“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).
Sign in to add a benchmark score for this model.
Darwin-180B-RSI launches
Request a demo or quote from Darwin-180B-RSI
Reviews
Write the first review of Darwin-180B-RSI
Used it? Your experience helps other buyers decide.
Write a reviewQuestions & answers
No questions yet. Be the first to ask about Darwin-180B-RSI.
Darwin-180B-RSI alternatives
Similar listings buyers compare against this one.
Microsoft AI's real-time speech-to-text model for voice agents, transcribing 60 languages with continuous language detection
Black Forest Labs' image generation and editing model with bounding-box layouts, multiple reference images and 4K output
Liquid AI's open-weight decision model, giving yes or no, choice or score answers from text and images
Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0
Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks
Microsoft AI's text-to-speech model for keeping one voice across 23 languages, with voice cloning from short audio
Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- Write a reviewHelp other buyers decide
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- ArticlesGuides, comparisons and research from the editorial desk and the community.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.