Darwin-180B-RSI
A 180B-parameter reasoning model released by the FINAL-Bench organization, showing strong performance on knowledge reasoning and mathematics benchmarks.
Timeline (last 90 days)
FINAL-Bench's 11 first places come from its Darwin-180B-RSI series models, covering areas such as MMLU-Pro (88.12), gpqa (94.44), MMMU_Pro (79.48), the LEXam series, and hmmt_feb_2026 (a score of 100).
Korean AI startup VIDRAFT released the 180B-parameter open-weight model Darwin-180B-RSI, hosted in the FINAL-Bench/Darwin-180B-RSI repository on Hugging Face.
As of October 19, 2026, Darwin-180B-RSI ranks first on the AIhoster leaderboard for MMLU-Pro at 88.1% and first for GPQA Diamond at 94.4%.
Darwin-180B-RSI used a thinking budget of 131,072 tokens in both the MMLU-Pro and GPQA Diamond evaluations.
Darwin-180B-RSI's MMLU-Pro evaluation used single sampling (temperature=1.0, top_p=0.95, top_k=20), and the GPQA Diamond evaluation used majority voting over up to 16 samples with the same parameters.
The top five in MMLU-Pro are separated by only 0.6 percentage points: Darwin-180B-RSI 88.1%, MiniMax-M2.1 88.0%, Intern-S2-Preview 88.0%, Qwen3.5-397B-A17B 87.8%, DeepSeek-V4-Pro 87.5%.
Under the AIhoster methodology, Darwin-180B-RSI's scores come from VIDRAFT's own model card report and have not been independently reproduced by a third party.
The materials do not specify Darwin-180B-RSI's license terms, architecture details, training hyperparameters, or dataset information, and the actual weight downloadability and inference service availability are also unconfirmed.
GPQA Diamond leaderboard top five: Darwin-180B-RSI 94.4% (180B), Kimi-K3 93.5% (2.8T), Darwin-397B-ZTC 93.4% (404B), Ornith-1.5-397B 92.8% (403B), Qwen3.8-2.4T-A95B 92.6% (2.4T).
Darwin-397B-ZTC's 93.4% on GPQA Diamond used greedy decoding with no majority voting, which is not under the same inference budget as Darwin-180B-RSI's 16-sample majority voting.
3 more timeline events
All Facts (13)
FINAL-Bench's 11 first places come from its Darwin-180B-RSI series models, covering areas such as MMLU-Pro (88.12), gpqa (94.44), MMMU_Pro (79.48), the LEXam series, and hmmt_feb_2026 (a score of 100).
50%UnverifiedThe top five in MMLU-Pro are separated by only 0.6 percentage points: Darwin-180B-RSI 88.1%, MiniMax-M2.1 88.0%, Intern-S2-Preview 88.0%, Qwen3.5-397B-A17B 87.8%, DeepSeek-V4-Pro 87.5%.
50%UnverifiedVIDRAFT's Darwin product line has multiple models at different scales and inference configurations near the top of the GPQA leaderboard, including 180B-RSI (first), 397B-ZTC (third), and 398B-JGOS (ninth, 90.9%), indicating a multi-version parallel iteration pipeline.
50%UnverifiedUnder the AIhoster methodology, Darwin-180B-RSI's scores come from VIDRAFT's own model card report and have not been independently reproduced by a third party.
50%UnverifiedThe materials do not specify Darwin-180B-RSI's license terms, architecture details, training hyperparameters, or dataset information, and the actual weight downloadability and inference service availability are also unconfirmed.
50%UnverifiedAs of October 19, 2026, Darwin-180B-RSI ranks first on the AIhoster leaderboard for MMLU-Pro at 88.1% and first for GPQA Diamond at 94.4%.
50%UnverifiedDarwin-180B-RSI used a thinking budget of 131,072 tokens in both the MMLU-Pro and GPQA Diamond evaluations.
50%UnverifiedDarwin-180B-RSI's MMLU-Pro evaluation used single sampling (temperature=1.0, top_p=0.95, top_k=20), and the GPQA Diamond evaluation used majority voting over up to 16 samples with the same parameters.
50%UnverifiedGPQA Diamond leaderboard top five: Darwin-180B-RSI 94.4% (180B), Kimi-K3 93.5% (2.8T), Darwin-397B-ZTC 93.4% (404B), Ornith-1.5-397B 92.8% (403B), Qwen3.8-2.4T-A95B 92.6% (2.4T).
50%UnverifiedDarwin-397B-ZTC's 93.4% on GPQA Diamond used greedy decoding with no majority voting, which is not under the same inference budget as Darwin-180B-RSI's 16-sample majority voting.
50%UnverifiedThe top of MMMU_Pro is Darwin-180B-RSI with a score of 79.48; the top of ScreenSpot-Pro is meta-models/Muse-Glimmer-30B with a score of 75.4; and the first place on open-asr-leaderboard is Edge0/ARK-ASR-3B with a score of 4.76 (lower is better).
50%UnverifiedThe Darwin model series has public records on Hugging Face; referring to VIDRAFT's previously released Darwin-180B-RSI, which once topped five Hugging Face benchmarks, one can view the published checkpoints on its organization homepage.
50%UnverifiedKorean AI startup VIDRAFT released the 180B-parameter open-weight model Darwin-180B-RSI, hosted in the FINAL-Bench/Darwin-180B-RSI repository on Hugging Face.
50%