# K2 Horizon 375B A23B - BenchAudit True Index Profile

**Integrity Profile: Clean Generalization**

> The Verdict: Performs Better on Fresh Private Evals Than Legacy Tests. K2 Horizon 375B A23B (evaluated at high effort) outperforms its legacy benchmark standing (4.4p) when moved to private, leak-resistant evaluations (11.8p), posting a -4.7p Low-Tier Boost across 3 matched domains, particularly in Agentic & Enterprise (-7.8p boost) and Coding & SWE (-5.6p boost)—a strong indicator of genuine generalization.

- **Canonical URL:** https://benchaudit.org/models/k2-horizon-375b-a23b__high/
- **True Index Rank:** #42 of 44
- **Creator:** Institute of Foundation Models
- **Weights:** Open Weights
- **Clean Evals Avg (Tiers S & A):** 11.8p (6 evals)
- **Legacy Benchmarks Avg (Tiers C & D):** 4.4p (4 evals)
- **Low-Tier Boost (Matched Delta):** -4.7p
- **Updated:** 2026-11-22

## True Index: Domain Rankings

| Capability Domain | Sub-Rank | Clean Evals (S/A) Avg | Top-Tier Evals |
| --- | ---: | ---: | ---: |
| Agentic & Enterprise | #41 | 11.2p | 2 |
| Coding & SWE | #40 | 9.7p | 1 |
| Math & Scientific Reasoning | #43 | 7.9p | 2 |

## Head-to-Head: Clean Evals vs. Legacy Benchmarks

| Capability Domain | Clean Evals (Tiers S & A) | Legacy Benchmarks (Tiers C & D) | Public Lift (Boost) |
| --- | ---: | ---: | ---: |
| Agentic & Enterprise | 2 (11.2p) | 2 (3.4p) | -7.8p |
| Coding & SWE | 1 (9.7p) | 1 (4.1p) | -5.6p |
| Math & Scientific Reasoning | 2 (7.9p) | 1 (5.7p) | -2.2p |

## Audited Benchmark Performance of K2 Horizon 375B A23B

| Benchmark | Benchmark Integrity Status | Domain | Rank | Percentile |
| --- | --- | --- | ---: | ---: |
| [AA-Omniscience v1.1](https://benchaudit.org/benchmarks/aa-omniscience-v1-1/) | Private / Fresh | General Knowledge & QA | #39/47 | 18.5p |
| [AA-Briefcase v1.1](https://benchaudit.org/benchmarks/aa-briefcase-v1-1/) | Private / Fresh | Agentic & Enterprise | #42/50 | 16.3p |
| [MLCR-AA](https://benchaudit.org/benchmarks/mlcr-aa/) | Private / Fresh | Math & Scientific Reasoning | #41/46 | 15.8p |
| [FrontierCode-AA 2.0](https://benchaudit.org/benchmarks/frontiercode-aa-2-0/) | Private / Fresh | Coding & SWE | #42/46 | 9.7p |
| [GDPval-AA v2.1](https://benchaudit.org/benchmarks/gdpval-aa-v2-1/) | Public / Aging | Agentic & Enterprise | #39/43 | 6.8p |
| [AutomationBench-AA](https://benchaudit.org/benchmarks/automationbench-aa/) | Private / Fresh | Agentic & Enterprise | #47/50 | 6.1p |
| [Terminal-Bench-Science 0.1](https://benchaudit.org/benchmarks/terminal-bench-science-0-1/) | Public / Aging | Math & Scientific Reasoning | #40/41 | 5.7p |
| [Terminal-Bench 4.0](https://benchaudit.org/benchmarks/terminal-bench-4-0/) | Public / Aging | Coding & SWE | #48/50 | 4.1p |
| [HLE-Diamond](https://benchaudit.org/benchmarks/hle-diamond/) | Semi-Exposed | General Knowledge & QA | #48/49 | 2.1p |
| [CritPt v2](https://benchaudit.org/benchmarks/critpt-v2/) | Private / Fresh | Math & Scientific Reasoning | #39/40 | 0.0p |
| [GDP.pdf](https://benchaudit.org/benchmarks/gdp-pdf/) | Public / Aging | Agentic & Enterprise | #50/50 | 0.0p |
