November 14 | IBM Vienna
Efficient Local Agents
Small model. Smart harness. Every team gets the same small language model and budget, plus Jev 1.14 by TypeSafe AI for typed decisions. The best harness wins.
Apply Now- Date
- Saturday, November 14, 2026, 09:00 – 19:30+
- Location
- IBM Vienna, Obere Donaustraße 95
- Seats
- About 30, by application. Teams of up to 4, or join solo
- Apply by
- Saturday, November 7, 23:59 CET
The challenge
No bigger model to hide behind
Frontier models make a lot of agents look good. We want to see what great engineering gets out of a small one. The challenge itself is revealed at the opening, and coding assistants are welcome while you build.
- Every team gets the same small language model and the same budget.
- Jev 1.14 by TypeSafe AI, a System One model, adds fast typed decisions like "supported or not?" and "retry or continue?".
- You build the harness: planning, tools, memory, retrieval and verification.
- Test against our live benchmark all day, with instant feedback per task.
- At 17:00 your agent runs once, on its own, on an unseen final scenario.
- Scoring is 50% accuracy and 50% cost.
Schedule
- 09:00Doors open
- 09:15Opening & challenge reveal
- 12:00Pizza lunch
- 17:00Final evaluation
- 18:00Leaderboard reveal, solution presentations & award ceremony
- 19:30+Drinks, food & networking
Subject to change. Organized by AIM in collaboration with IBM, Klartext AI and Texterous.
Frequently Asked Questions
How do I get a seat?
Seats are limited to about 30 and given out by application. Apply with your CV or LinkedIn profile by November 7, 2026, 23:59 CET. Confirmations are sent by November 9.
Can I join without a team?
Yes. Teams have up to 4 people. Everyone applies on their own: enter the same team name as your teammates, or write "alone" or "assign" if you want to work solo or be placed in a team.
Which models can we use?
At runtime, only the two provided models: the small language model, announced at the opening, and Jev 1.14 by TypeSafe AI. Coding assistants are welcome while you build.
How are agents scored?
At 17:00 you freeze your harness and your agent runs once, on its own, on an unseen final scenario. The score is 50% accuracy (right answers and right sources) and 50% cost.
What should I bring?
Your laptop with an IDE and dev tools set up. IBM emails Wi-Fi access links, and food is provided all day.
What can I win?
IBM credits and goodies for the top teams, and certificates for winners and all participants.