K2 Horizon 32B

K2-Horizon-32B

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

Pricing

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

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

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

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