# K2-Horizon-32B ![Organization](https://img.shields.io/badge/Org-IFM-555) ![Context](https://img.shields.io/badge/Context-512K%20tokens-1f6feb) ![Pricing](https://img.shields.io/badge/Price-Open%20weights-2da44e) ![License](https://img.shields.io/badge/License-Apache%202.0-d29922) ![Modality](https://img.shields.io/badge/Modality-Text-8957e5) ![Released](https://img.shields.io/badge/Released-Sep%202026-555) > **Quick answer:** K2-Horizon-32B is IFM's large dense model in the K2-Horizon family — 32B parameters, all activated, with a native 512K-token context window. This is the Stage 1 training checkpoint; results are for stage 1 only, with stage 2 results expected later. Open weights, Apache 2.0. ## At a Glance **Where K2-Horizon-32B leads** - Evaluated on the same agentic, coding, and reasoning benchmark suite as the rest of the K2-Horizon family for direct comparison - 512K native context window - Fully open weights **Where it lags** - Trails Qwen3.8-27B substantially on τ³-Banking, Terminal-Bench 2.1, and GPQA Diamond in IFM's own comparison table - This is a Stage 1 checkpoint — final (Stage 2) results are not yet published, so current numbers may understate the finished model **Best for:** teams wanting to track K2-Horizon's dense-model trajectory before the Stage 2 checkpoint lands; not yet the strongest dense option in its size class. ## What K2-Horizon-32B Is K2-Horizon-32B is the large dense member of IFM's K2-Horizon family — a 32B-parameter decoder-only model evaluated on the same benchmark suite as its MoE siblings for apples-to-apples comparison. IFM explicitly labels current published results as "Stage 1" of the final model's training, with Stage 2 results to follow — a transparency choice that lets outside observers track capability changes across a checkpoint's training rather than judging only a finished model. ## Specifications | Field | Value | |---|---| | Organization | IFM | | Parameters | 32B (dense, all activated) | | Context window | 524,288 tokens (512K) | | Architecture | Dense decoder-only | | License | Apache 2.0 | | Release date | 2026-09 (Stage 1 checkpoint) | | Modality | Text | ## Pricing | | Input (per 1M tokens) | Output (per 1M tokens) | |---|---|---| | Open weights | — | — | Open weights: free to download and self-host. Validated SGLang recipe: 2x H200, TP=2. ## Context Window K2-Horizon-32B has a **524,288-token (512K) context window**, native from the midtraining stages onward. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [τ³ Banking](/benchmarks/tau-bench/tau3-banking) | 22.5 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-32B) | 2026-09 | | [TerminalBench 2.1](/benchmarks/tbench/terminalbench-21) | 36.6 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-32B) | 2026-09 | | [SciCode](/benchmarks/scicode-team/scicode) | 30.2 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-32B) | 2026-09 | | [Humanity's Last Exam](/benchmarks/scale-ai/humanitys-last-exam) (no tools) | 22.8 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-32B) | 2026-09 | | [GPQA Diamond](/benchmarks/rein-et-al/gpqa-diamond) | 82.3 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-32B) | 2026-09 | | [CritPt](/benchmarks/critpt-team/critpt) | 1.4 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-32B) | 2026-09 | | [AA-LCR](/benchmarks/artificial-analysis/aa-lcr) | 65.3 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-32B) | 2026-09 | *AA-Omniscience Accuracy (16.8) and Non-Hallucination (58.3) were also reported but use a different metric structure than the platform's combined AA-Omniscience Index — pending mapping decision, not yet added. Scores above are reported by IFM (Stage 1 checkpoint) and shown for context; not Benchgen measurements.* ## Use K2-Horizon-32B via API ```python from openai import OpenAI client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY") response = client.chat.completions.create( model="IFM/K2-Horizon-32B", messages=[{"role": "user", "content": "Explain the result step by step."}], temperature=1.0, top_p=0.95, extra_body={"chat_template_kwargs": {"reasoning_effort": "high"}}, ) print(response.choices[0].message.content) ``` ## Frequently Asked Questions

What is K2-Horizon-32B?

It's IFM's large dense model in the K2-Horizon family: 32B parameters, 512K-token context. Current results reflect a Stage 1 training checkpoint.

What is K2-Horizon-32B's context window?

524,288 tokens (512K).

How much does K2-Horizon-32B cost?

Open weights — free to download; cost is hosting/inference only.

Is K2-Horizon-32B open source?

Yes, released under the Apache 2.0 license.

--- *Specs and scores sourced from IFM's official Hugging Face model card; third-party benchmark scores attributed inline. Last updated 2026-09-03.*

K2 Horizon 32B — Training jobs

Training jobs