Inception releases the decision-making model Mercury Decide, claiming it can make up to 14 decisions per second, but the core benchmark result has not been verified by a public leaderboard.
Article
Inception releases Mercury Decide: up to decisions per second 14 times, officially claiming benchmark performance surpasses Jev
Beating AI News reports that AI company Inception has released the decision-making model Mercury Decide, which performs the same type of task as Jev. It does not generate long-form text; instead, it directly makes choices, judgments, and scores, and returns the probability of each answer, making it suitable for Agent selecting tools, routing, or controlling workflows.
Inception says Mercury Decide is OpenRouter the best-performing decision-making model on JevBench v1.4, completing up to 14 decisions per second. However, the company has not published a specific benchmark score, and Mercury Decide has not yet appeared on JevBench's public leaderboard, so this result has so far been reported only by Inception itself.
Mercury Decide is compatible with the /v1/systemone interface used by Jev. Applications already integrated with TypeSafe SDK can call it directly. The model supports a maximum of 32768 tokens context.
Mercury Decide is currently available for free on OpenRouter Early Access. Inception also showcased a demo of it playing chess against Jev.
Original link https://m.theblockbeats.info/flash/369915
Key points
01
Inception has released the decision-making model Mercury Decide.
02
The original text states that Mercury Decide does not generate long-form text, but directly makes choices, judgments, and scores, and returns the probability of each answer.
03
The original text states that the model can help Agent select tools, perform routing, or control workflows.
04
Inception says Mercury Decide is OpenRouter the best-performing decision-making model on 14 JevBench v1.4, completing up to decisions per second.
05
The company has not published a specific benchmark score, and Mercury Decide has not yet appeared on JevBench's public leaderboard.
06
Mercury Decide is compatible with the /v1/systemone interface used by Jev.
07
The original text states that applications integrated with TypeSafe SDK can call Mercury Decide directly.
08
The model supports a maximum of 32768 tokens context.
09
Mercury Decide is currently available for free on OpenRouter Early Access。
10
Inception showcased a demo of Mercury Decide playing chess against Jev.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
Mercury Decide is a model specialized in decision-making: given a question or candidate options as input, it directly provides a choice, judgment, score, and the probability of each answer, rather than generating long passages of text. The uses described in the original text include helping Agent select tools, assign requests, and control workflows.
Why it matters to readers
If the speed, interface compatibility, and decision-making capabilities described in the original text hold, developers can integrate it into Agent applications to handle tool selection, routing, and process control. However, the claim of benchmark leadership has so far been published only by Inception.
Beginners can think of it as a Agent“selector” in which it does not provide long-form answers, but helps programs determine which tool to call next or which workflow to follow. It can currently be tried for free through OpenRouter Early Access.
Risks and unknowns
Inception has not published a specific benchmark score for JevBench v1.4.
Mercury Decide has not yet appeared on JevBench's public leaderboard, so the claim that it is the “best-performing” model lacks public verification.
The original text does not specify the duration of the free Early Access period, usage limits, or subsequent pricing.
The original text provides no third-party verification of the speed, interface compatibility, context length, or actual results.
Related Developments
Loading event timeline…
The chess-playing demo alone cannot prove the model's performance on other decision-making tasks.
Related concepts
Decision model
This term is not in the glossary yet. Browse related concepts in the glossary.
