Model library / IFM
K2-Horizon-32B-Stage1 is a 32-billion-parameter dense decoder-only model released by IFM for benchmarking agentic, coding and reasoning tasks.
Parameters in total
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
K2-Horizon-32B-Stage1 is a 32-billion-parameter decoder-only language model released by IFM. It is presented as a strong dense baseline and has been evaluated on agentic, coding and reasoning benchmarks during its first training stage. The model uses the K2HorizonForCausalLM architecture without expert layers and is intended for research and development purposes.
Deploying K2-Horizon-32B-Stage1 requires handling a 34.8-billion-parameter model stored in bfloat16 format. Its native context window spans 524,288 tokens, enabling very long inputs. The model is released under the Apache-2.0 licence, allowing commercial and non-commercial use. Absence of mixture-of-experts layers simplifies inference, but the size demands substantial GPU memory and bandwidth.
Specification
| Repository | IFM/K2-Horizon-32B |
| Publisher | IFM |
| Published | 2 September 2026 |
| Architecture | K2HorizonForCausalLM |
| Model type | k2_horizon |
| Parameters | 34,779,304,960 |
| Layers | 64 |
| Hidden size | 5,120 |
| Attention heads | 64, 8 key/value heads |
| Context length | 524,288 tokens |
| Vocabulary | 250,624 tokens |
| Weight format | bfloat16 |
| Licence | apache-2.0 |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 64.8 GiB | Full precision as released |
| fp8 | 32.4 GiB | 8-bit, near-lossless on most models |
| int4 | 16.2 GiB | 4-bit, smallest footprint |
The cache costs 256 KiB per token at 16-bit, so the full 524,288-token context of one request needs about 128.0 GiB. Concurrency multiplies that number, not the weights.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 4 | 11.6 GiB | about 47,589 tokens |
| RTX 5090 32GB | 4 | 40.4 GiB | about 165,554 tokens |
| L40S 48GB | 2 | 16.6 GiB | about 68,069 tokens |
| A100 80GB | 1 | 4.7 GiB | about 19,326 tokens |
| H100 80GB | 1 | 4.7 GiB | about 19,326 tokens |
| RTX PRO 6000 96GB | 1 | 19.1 GiB | about 78,309 tokens |
| H200 141GB | 1 | 59.6 GiB | about 244,197 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve IFM/K2-Horizon-32B \ --max-model-len 45056 \ --gpu-memory-utilization 0.90
Questions
The weights take about 65 GiB at the released precision. That fits on one A100 80GB, which leaves roughly 5 GiB for the key-value cache.
The repository declares apache-2.0. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 524,288 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email [email protected] with the model name and roughly what volume you expect.
Also in the library
K2-Horizon-375B-A23B is an open-weight sparse MiE language model from IFM with 379 B parameters, 23 B active per token and a 512 …
K2-Horizon-7B is a 9-billion-parameter decoder-only language model with a 524k token context window, released under Apache-2.0.
K2-Horizon-3.7B is a 3.7 billion-parameter dense decoder-only model with a 512 K token context, released openly by IFM.
K2-Horizon-MoVA-36B-A4B is a sparse Mixture-of-Experts language model with MoVA attention, released by IFM.
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.