K2 Horizon 375B A23B
K2 Horizon 375B A23B by the MBZUAI Institute of Foundation Models is a sparse MoE flagship with 23B active parameters for complex reasoning, coding and
Model Overview
Capabilities, design details, and architectural traits
K2 Horizon 375B-A23B - the enterprise powerhouse of the Horizon fleet
K2 Horizon 375B-A23B is a sparse mixture-of-experts flagship built for demanding enterprise workloads, and it anchors a fleet that spans from 0.9B edge models up to this deployment target.
| Trait | Detail |
|---|---|
| Architecture | Sparse MoE with 375 billion total parameters and roughly 23 billion active per token, drawing on the capacity of a much larger model without using every parameter for every token |
| Positioning | Described by IFM as the fleet's enterprise powerhouse |
| Target workloads | Complex reasoning, software engineering, research and long-horizon agentic tasks |
| Openness | Released under Apache 2.0 as part of what IFM calls the first fully open model fleet for agents, exposing the complete development process through agentic post-training |
| Released artifacts | Intermediate checkpoints, training data or detailed data-construction recipes, training code, configurations, fine-grained logs, evaluation results and final weights |
| Deployment profile | A multi-accelerator, cluster-scale deployment rather than a single-device model |
One connected fleet
The 375B-A23B model shares core architecture, vocabulary, training methodology, interfaces, evaluation infrastructure and deployment tooling with its five fleet siblings. This consistency is designed to make it easier to move between sizes, route work dynamically and study capability across scale, so teams can prototype on a smaller model and scale up to the flagship.
Reasoning built into pretraining
Across the fleet, IFM incorporated reasoning directly into pretraining, with nearly 17% of the pretraining corpus consisting of explicit reasoning trajectories. The open release is intended to let researchers study how reasoning, tool use, planning and agentic capabilities emerge, and to reproduce or adapt the methods rather than treat the final checkpoint as an opaque starting point.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | MBZUAI Institute of Foundation Models | Anthropic | Anthropic |
| Release Date | September 3, 2026 | September 1, 2026 | July 24, 2026 |
| Knowledge Cutoff | - | - | May 2026 |
| Context & Limits | |||
| Context Window | 524K | - | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $10 | $5 |
| Output Pricing | Free Best Output Pricing | $50 | $25 |
| Modalities | |||
| Inputs | text | textimagefile | textimage |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 47.3 | 65.7 Best Intelligence Index | 63.1 |
| Coding Index | 61.5 | 81.6 Best Coding Index | 78.0 |
| Agentic Index | 42.8 | 61.3 Best Agentic Index | 59.2 |
GPQA Benchmark
Graduate-level reasoning and expert Q&A evaluation.
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.