Liquid AI releases d1-ultra, ranks first on the Artificial Analysis decision-making leaderboard

AT A GLANCE

The report states that d1-ultra topped the decision-making leaderboard with 65 points, but its accuracy on knowledge questions was about 32%, and it selected "I don't know" more often.

Article

Liquid AI New model d1-ultra tops the decision-making leaderboard: surpasses Jev, but accuracy on knowledge questions falls to 32%Beating AI Flash, Liquid AI releases multimodal decision-making model d1-ultra, scoring Decision Index v1.0 on Artificial Analysis's 65 leaderboard, surpassing Jev 1.14's 57 points to rank first.

Decision-making models are not responsible for chatting or generating articles. Instead, they read text and images, directly determine which option is more likely to be correct, and provide probabilities. For example, they can inspect circuit-board defects, verify product images against descriptions, or review customer-submitted claim photos.

However, there is a contrast behind the new model's top ranking. Among Artificial Analysis's 500 knowledge test questions,d1-ultra answered only about 32%correctly, below the previous generation d1's 41%. Its advantage is making fewer unfounded answers. Among questions it did not answer correctly, the proportion selecting "I don't know" increased from 2%to 85%. When faced with uncertain questions, it is more willing to admit it does not know rather than risk giving a wrong answer.

This also explains why its ranking improved substantially. Artificial Analysis assigns 12.5%weight each to knowledge accuracy and the ability to avoid incorrect answers; the remaining score comes from complex reasoning, judgment in tool use, and safety classification.

Liquid AI's own 112 evaluations also show that d1-ultra outperformed Jev in all task categories. 1.14。

d1-ultra is available through Liquid API and Vercel, while model weights have not yet been released. The charge per million input Token is 0.042 USD,d1 times that of the previous generation 2.1, with no output charge. If the same image is used to answer multiple questions, the image input cost is recalculated for each question.

Original link https://m.theblockbeats.info/flash/372287

Key points

Beating AI says that Liquid AI released the multimodal decision-making model d1-ultra。
The report states that d1-ultra scored Decision Index v1.0 on Artificial Analysis's 65 leaderboard, higher than Jev 1.14's 57 points, ranking first.
Among Artificial Analysis's 500 knowledge test questions,d1-ultra's accuracy was about 32%, below the previous generation d1's 41%。
Among questions it did not answer correctly,d1-ultra's proportion of selecting "I don't know" was 85%, versus d1 for the previous generation 2%。
The report states that Artificial Analysis assigns 12.5%weight each to knowledge accuracy and the ability to avoid incorrect answers.
Liquid AI's independently conducted 112 evaluations show that d1-ultra outperformed Jev in all task categories. 1.14。
The report states that d1-ultra is available through Liquid API and Vercel, but model weights have not yet been released.
The price per million input Token is 0.042 USD,d1 times that of the previous generation 2.1, with no output charge.
When the same image is used to answer multiple questions, the image input cost is recalculated for each question.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

These models primarily read text or images and make judgments among candidate options while providing probabilities. The d1-ultra in the report answered fewer knowledge questions correctly, but selected "I don't know" more often when facing uncertain questions rather than continuing to answer. In addition to knowledge accuracy, the leaderboard also calculates performance in avoiding incorrect answers, complex reasoning, judgment in tool use, and safety classification.

Why it matters to readers

If the relevant evaluations can be independently reproduced, a greater tendency to refuse may suit review scenarios that prioritize reducing incorrect judgments; however, practical value also depends on performance on specific tasks and input costs.

Beginners should consider accuracy, refusal rate, evaluation weights, and pricing together, and should not judge whether a model is suitable for their task solely based on a "first-place ranking".

Risks and unknowns

The original text does not provide a primary announcement link from Liquid AI or Artificial Analysis.
Liquid AI The 112 evaluations are vendor self-assessments, and the text does not state whether they were independently verified.
The text does not provide Decision Index v1.0’s complete scoring criteria and all results.
It remains unclear whether third parties can reproduce the knowledge accuracy, refusal rate, and leaderboard advantage.

Liquid AI releases d1-ultra, ranks first on the Artificial Analysis decision-making leaderboard

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center