Upright launches the first consequential decision model for AI agents

Upright Project, a Finnish technology company, today launched Eevo, the first consequential decision model: a new type of AI model designed to help AI agents make better decisions by weighing what each option leads to, and for whom.

As AI agents gain autonomy, instructions cannot anticipate every situation. An agent may need to decide whether to roll back a software release, issue an exceptional refund or wake an engineer in the middle of the night. Eevo helps agents weigh what each option means for the people and organisations affected before acting, giving each option a calibrated probability.

Decision models are a fast-growing class of AI models that answer defined questions with calibrated probabilities rather than generating free-form text. Eevo is the first decision model developed in Europe. It opens a new category of the consequential decision model that connects the rise of decision models to AI alignment: agents that account for what follows from their decisions before they make them, efficiently enough to be commercially practical.

Annu Nieminen, Co-Founder & CEO of Upright, said:

"AI alignment is dangerously falling behind capability progress, and very little innovation has been introduced. Eevo is a rare concrete tool launch in the space. It gives people a practical way to optimise AI decisions with the interest of multiple stakeholders, while making priorities and trade-offs transparent.”

Eevo outperforms leading US decision models in multiple public benchmarks

In independent public benchmarks, Eevo-1 scored 72.8% on STEER, a test of economic rationality with 16,300 questions. That is ahead of TypeSafe's Jev (70.3%), Microsoft-Decision-1 (66.4%) and OpenAI Decisions (64.8%). It also led the field on PokerBench, which tests strategic reasoning with incomplete information, with 87.7% against 81.0% for OpenAI Decisions.

On Upright's own evaluations of expected-value reasoning, game theory and risk in sequential decisions, it scored 81.9%, 84.2% and 83.3% respectively, ahead of every model it was compared with. On CoCoCo, Upright's benchmark for how consistently a model sizes consequences against each other, Eevo-1 scored 84.2. The best general-purpose frontier model tested scored 79.7, at twelve times the cost.

Juho Ojala, Co-Founder & CTO at Upright, said:

“Real-world decisions require comparing things that don't naturally share a scale: a lost sale, an hour of an engineer's sleep and a customer's trust. We trained Eevo to compare consequences in pairs and keep those comparisons consistent across a whole decision. That consistency is what makes an expected value worth acting on.”

How Eevo works

Eevo can handle judgement calls inside AI agents, support human decision-making across a wide range of business contexts, and power high-volume automated decisions in software. It is available for anyone to try through an API, an MCP server for agents such as Claude, ChatGPT and Codex, and the Eevo console. New accounts include free decisions, while Enterprise customers can deploy privately on AWS, Google Cloud and Azure.

The name Eevo comes from expected value (EV): the probability-weighted sum of an option's outcomes, which is the quantity the model estimates.

Developed by a Finnish team

Eevo is developed by Upright's Finnish team of around 30 people. Co-founded and led since 2017 by Annu Nieminen, the company is one of the only AI model developers led by a woman. Upright has raised €5 million funding from Planet A and First Fellow Partners in 2022.

Upright has previously built models that quantify the positive and negative impacts of companies across 800 impact categories, used by financial institutions managing over €1 trillion in assets. Eevo now brings the same core technology – making different things comparable – to AI agents' decisions.

Media contact:

Pauliina Bogdanoff, pauliina.bogdanoff@uprightproject.com, +358440241616

More information & trial:

https://www.uprightproject.com/eevo/

Images: here

About Upright

The Upright Project is a Helsinki-based, independent, AI-native company that builds models to quantify consequences. Its impact model of the global private sector quantifies the sustainability impacts, risks and opportunities of products, companies and funds, and is used by more than 300 businesses and investors. Eevo, launched in October 2026, extends the same consequence-sizing technology to decisions made by AI agents and people.