0

Hypothesis Forge

Frontier

Novelty-aware, evidence-grounded research discovery environment

Open
Domain
rl-env
License
apache-2.0
Published
Mar 2026
Canonical
app.primeintellect.ai/dashboard/environments/casella/hypothesis-forge

Cite

Notes

Only stored in your browser.

Attribution

Leaderboard scores
prime-hub
Attribution policy →

Top score 48.9% by GPT-5 - 3 models reporting (3 frontier)

Score history

3
35%51%68%84%100%Apr 25May 25Jun 25Jul 25Aug 25GPT-4.1 MiniGPT-5 MiniGPT-5

Top models

3
Hypothesis ForgeBar chart with 3 bars. Highest value: GPT-5 at 48.9.
3 models

Related tools

1
View all

Implementations, trainers, datasets and scaffolds linked to this eval.

Hypothesis Forge RL Env (Community)

Official

Novelty-aware, evidence-grounded research discovery environment

ImplementationRL EnvTool UseResearchCreative

FAQ

What is Hypothesis Forge?
Novelty-aware, evidence-grounded research discovery environment
What is the current top score on Hypothesis Forge?
The top reported score is 48.9% by GPT-5, across 3 models reporting (3 from frontier labs).
How can a model improve its Hypothesis Forge score?
Tools linked to Hypothesis Forge on Sophon include Hypothesis Forge RL Env (Community) - RL environments, datasets, and scaffolds that target this eval.
What license is Hypothesis Forge under?
Hypothesis Forge is available under apache-2.0.