Preferences and analytics

BenchSift keeps your language, theme, and comparison preferences in your browser. Audience measurements also help improve the site.

Back

Institute of Foundation Models

K2 Horizon 3.7B

16.2 Intelligence Index2026-09-03ReasoningOpen weights
Artificial Analysis Indices
Default
Intelligence16.2
Coding26.1
Agentic18.4
Standard Benchmarks
GPQA Diamond69.2%
HLE13.9%
SciCode22.0%
AA-LCR62.3%
Terminal-Bench 2.128.1%
τ-Bench Banking19.6%
Capabilities
Context window524K tokens
Openness Index53 / 100
Input
text
Output
text
Meta-information
Tokens used (verbosity)180.0M