K2 Horizon 3.7B
K2 Horizon 3.7B by the MBZUAI Institute of Foundation Models is a dense 3.7B model with 512K context, adjustable reasoning effort, and a fully open training
Model Overview
Capabilities, design details, and architectural traits
K2 Horizon 3.7B - the fully open small model with 512K context
K2 Horizon 3.7B is the small dense member of the K2 Horizon fleet, a connected family of six models from 0.9B to 375B released by MBZUAI's Institute of Foundation Models. It is positioned as a strong small-model baseline and targeted at phones and other on-device applications.
What sets it apart
| Trait | Detail |
|---|---|
| Fleet position | Dense 3.7B member of the six-model K2 Horizon fleet, sharing architecture, vocabulary, and tooling with sizes from 0.9B to 375B |
| Context window | Native 524,288-token context from midtraining stages onward |
| Full lifecycle openness | Weights, training data and recipe, training code, and intermediate checkpoints are public under Apache 2.0 |
| Adjustable reasoning depth | Reasoning effort is selected per request via chat template kwargs |
| Flexible tool calling | Supports json, xml, and xml_typed tool call formats, with xml as the default |
One connected fleet
The six K2 Horizon models share core architecture, vocabulary, training methodology, interfaces, and deployment tooling. This consistency is designed to make it easy to move between sizes, route work dynamically, and study capability and efficiency across scale. The 3.7B is one of the sizes IFM ships fully open today, with data, recipe, code, and intermediate checkpoints already released.
Running it well
IFM's guidance is explicit: use reasoning effort high, temperature 1.0, top_p 0.95, and at least 32,768 output tokens, since truncated reasoning counts as a failed response. The model is served through vLLM and SGLang with dedicated k2_horizon reasoning and tool-call parsers, and GGUF builds are available for llama.cpp.
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 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 524K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $4 | $2 |
| Output Pricing | Free Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 15.6 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 26.1 | - | - |
| Agentic Index | 16.5 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from MBZUAI Institute of Foundation Models
Other models by MBZUAI Institute of Foundation Models
K2 Horizon 375B A23B
K2 Horizon MoVA 36B A4B
K2 Horizon 7B
K2 Horizon 0.9B
Top AI Models
Leading alternatives by intelligence score
