K2 Horizon 32B

Institute of Foundation Models

K2 Horizon 32B is the large dense member of IFM's (MBZUAI) fully open K2 Horizon family: a 32B-class decoder-only reasoning model (64 layers, 250K vocabulary; about 34.8B parameters in the released safetensors) pretrained on 22.9T tokens, with a native 524,288-token (512K) context window from midtraining onward. Text in, text out, with per-request reasoning effort (high recommended) and JSON/XML tool-call formats. Released 2026-09-03 under Apache 2.0; the main branch was updated on 2026-10-01 to a later SFT checkpoint and its model-card scores were refreshed on 2026-10-07 (they are higher than the launch-blog figures). BF16 weights plus FP8, NVFP4, and GGUF builds, with intermediate checkpoints, training data, code, and logs. No priced provider is cataloged yet.

Key Specifications

Parameters
32.0B
Context
-
Release Date
September 3, 2026
Average Score
43.5%

Timeline

Key dates in the model's history
Announcement
September 3, 2026
Last Update
October 31, 2026
Today
November 20, 2026

Technical Specifications

Parameters
32.0B
Training Tokens
22.9T tokens
Knowledge Cutoff
-
Family
-
Capabilities
MultimodalZeroEval

Benchmark Results

Model performance metrics across various tests and benchmarks

Reasoning

Logical reasoning and analysis
GPQA
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95; GPQA Diamond • Self-reported
82.8%

Other Tests

Specialized benchmarks
AA-LCR
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95; long-context reasoning • Self-reported
68.0%
CritPT
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95 • Self-reported
2.9%
Humanity's Last Exam
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95; without tools • Self-reported
25.2%
OmniScience
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95; OmniScience accuracy • Self-reported
18.2%
OmniScience (non-hallucination rate)
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95 • Self-reported
69.7%
SciCode
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95 • Self-reported
37.9%
Tau3 Banking
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95 • Self-reported
24.1%
Terminal-Bench 2.1
IFM K2-Horizon-32B Hugging Face model card, final checkpoint; reasoning_effort=high, temperature=1.0, top_p=0.95 • Self-reported
62.6%

License & Metadata

License
apache_2_0
Announcement Date
September 3, 2026
Last Updated
October 31, 2026

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.