/Tech4h ago

FutureSim's updated forecasting benchmark finds Fable 5 and Opus 4.8 leading with just 20% accuracy

GPT 5.5 trailed the leaders with 15% accuracy.

549132.6K
Florian Brand
Maksym Andriushchenko#1207
Original post
Maksym Andriushchenko@maksym_andr#1207inTech

FutureSim has been accepted to the ICML 2026 AI Forecasting Workshop as an *oral*!

Additionally, check out the recent results of Fable 5 and Opus 4.8 that @nikhilchandak29 has recently added:

Shashwat Goel@ShashwatGoel7

FutureSim will be presented as an Oral at the ICML AI Forecasting Workshop.

Join our talk in Seoul to hear more about:

How to evaluate agent adaptation as new information arrives, in an environment based on real-world events as they happened over time

10:57 AM · Jun 11, 2026 · 846 Views
Sentiment

Users congratulated the authors on FutureSim's oral acceptance at the ICML 2026 AI Forecasting Workshop because it marks a notable research achievement.

Pos
100.0%
Neg
0.0%
1 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS935BOOKMARKS2LIKES31RETWEETS1REPLIES2
Maksym Andriushchenko@maksym_andr

Excited to share that we have 2 papers accepted as orals (top 5 out of 84 accepted papers) at the ICML 2026 AI Forecasting Workshop! - Allocation, Not Volume: Test-Time Compute for Agentic Forecasting - FutureSim: Replaying World Events to Evaluate Adaptive Agents

4hViews 935Likes 31Bookmarks 2
Maksym Andriushchenko@maksym_andr

Allocation, Not Volume: Test-Time Compute for Agentic Forecasting: https://openreview.net/forum?id=VF3SHjyHB5

Maksym Andriushchenko@maksym_andr

Excited to share that we have 2 papers accepted as orals (top 5 out of 84 accepted papers) at the ICML 2026 AI Forecasting Workshop! - Allocation, Not Volume: Test-Time Compute for Agentic Forecasting - FutureSim: Replaying World Events to Evaluate Adaptive Agents

4hViews 258Likes 2Bookmarks 0
Maksym Andriushchenko@maksym_andr

FutureSim: Replaying World Events to Evaluate Adaptive Agents: https://openreview.net/forum?id=s0F0l0Jl7e

(longer version: https://arxiv.org/abs/2605.15188)

Maksym Andriushchenko@maksym_andr

Allocation, Not Volume: Test-Time Compute for Agentic Forecasting: https://openreview.net/forum?id=VF3SHjyHB5

4hViews 267Likes 1Bookmarks 0
Florian Brand@xeophon

@maksym_andr @nikhilchandak29 congrats sashwat!!

Maksym Andriushchenko@maksym_andr

FutureSim has been accepted to the ICML 2026 AI Forecasting Workshop as an *oral*!

Additionally, check out the recent results of Fable 5 and Opus 4.8 that @nikhilchandak29 has recently added:

4hViews 218Likes 4Bookmarks 0
Maksym Andriushchenko@maksym_andr

All accepted papers are available here: https://forecasting-workshop.github.io/

Maksym Andriushchenko@maksym_andr

FutureSim: Replaying World Events to Evaluate Adaptive Agents: https://openreview.net/forum?id=s0F0l0Jl7e

(longer version: https://arxiv.org/abs/2605.15188)

4hViews 89Likes 0Bookmarks 0