Liquid AI releases decision model d1, saying its evaluation score surpasses Jev
BIBIBI
AT A GLANCE
Liquid AI released decision model d1, saying its comprehensive evaluation score surpassed Jev for the first time.
Article
Jev overtaken for the first time, Liquid AI releases decision model d1
Beating AI News Flash: cutting-edge edge-side AI R&D company Liquid AI has released its first self-developed decision model, d1. It does the same type of work as Jev: rather than generating long-form text, it directly makes choices, judgments, and scores, and can help Agent select tools, perform routing, and conduct reviews.
Liquid reran the evaluation according to Hugging Face's comprehensive evaluation of decision models, Decision Index 0.2.1. d1's overall score was 58.9, surpassing Jev's 1.13 across 57.9。5 major categories. d1 led in artistic judgment, language understanding, and retrieval classification; it was slightly below Jev in tool calling, while lagging by 8 points in knowledge reasoning. Liquid said d1 is the first model to surpass Jev in this evaluation.
d1 supports Jev's three decision modes: Choice selects one option from multiple choices, Noul determines true or false, and Score assigns a score according to criteria. A single request can also complete multiple judgments simultaneously, returning probabilities and structured results directly, without first generating text and then parsing it.
d1 is currently available through Liquid API; the free version is `d1:free`. The model weights and parameter scale have not yet been disclosed, and Liquid also plans to integrate it with OpenRouter.
Original link https://m.theblockbeats.info/flash/369763
Key points
01
Liquid AI released its first self-developed decision model, d1, for making choices, judgments, and scores.
02
The original article says d1 can be used to Agent select tools, perform routing, and conduct reviews.
03
Liquid reevaluated it according to Decision Index 0.2.1, with d1's overall score being 58.9，Jev 1.13 being 57.9。
04
Among 5 major categories, d1 led in artistic judgment, language understanding, and retrieval classification; it was slightly below Jev in tool calling and lagged by 8 points in knowledge reasoning.
05
d1 supports three decision modes—Choice, Noul, and Score—and can complete multiple judgments simultaneously while returning probabilities and structured results.
06
d1 is available through Liquid API; the free version is d1:free.
07
The model weights and parameter scale have not yet been disclosed, and Liquid also plans to integrate d1 with OpenRouter.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
d1 is not primarily designed to generate long-form text; instead, it directly makes a choice, a true-or-false judgment, or a score for a given task. The uses described in the original article include Agent selecting tools, assigning routes, and conducting reviews.
Why it matters to readers
If the capabilities and evaluation results in the original article are reliable, d1 could provide an option for Agent workflows requiring structured decisions, and could reduce the steps of generating text first and then parsing the result.
Beginners can think of d1 as a “decision-maker”: after receiving options or criteria as input, it directly returns a choice, judgment, score, or probability instead of writing a long answer.
Risks and unknowns
The evaluation was rerun by Liquid according to Decision Index 0.2.1; the original text provided no public leaderboard or evidence of independent retesting.
The claim that it was the “first to surpass Jev in this evaluation” comes from Liquid and cannot yet be independently confirmed from the original text.
The model weights and parameter scale have not yet been disclosed.
Integration with OpenRouter remains a plan; whether and when it will be completed is unknown.
The original text did not provide Liquid API’s specific usage restrictions or complete pricing information.
Related Developments
Loading event timeline…
Related concepts
Decision model
This term is not in the glossary yet. Browse related concepts in the glossary.
