The LLM benchmark aggregator.

K2 Horizon MoVA 36B A4B

Institute of Foundation Models · 36B · A4B

Strongest in Knowledge work (#37 of 178), weakest in General intelligence (#105 of 204). Above par in 8 of 14 scopes. Among the models it meets almost everywhere, it finishes behind GPT-5.6 Terra and Kimi K3 and ahead of Mimo V2.5 Pro and GPT-4.1.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Agents56.8−11.3#50/2682/5

Knowledge work58.8−14.8#37/1782/2

Tool use55.5−18.5—/81✱0/1

Core abilities49.7−17.7#103/2041/3

General intelligence49.6−20.5#105/2041/3

Long context47.4−17.2—/0✱✱

Reasoning49.4−17.5—/178✱0/4

Science49.1−14.4—/122✱0/1

Expert reasoning47.2−13.7—/0✱✱

Coding51.1−18—/165✱0/5

Agentic coding52.1−16.6—/157✱0/4

Code generation51.4−18.6—/77✱0/2

Knowledge50.3−13.2—/138✱0/2

Factuality50.4−8.6—/0✱✱

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for K2 Horizon MoVA 36B A4B, left for the other.

NVIDIA Nemotron 3 Nano 4Bwon 9 · lost 0

Mistral Medium 3.5won 9 · lost 0

Mistral Large 3won 9 · lost 0

GPT-4o Miniwon 10 · lost 0

GPT-4.1 Nanowon 10 · lost 0

Granite 4.2 3Bwon 11 · lost 0

K2 Horizon 32Bwon 12 · lost 0

Granite 4.2 30Bwon 13 · lost 0

Qwen3.6 35B A3Bwon 13 · lost 1

Nemotron 3 Superwon 12 · lost 1

Claude Haiku 4.5won 9 · lost 1

GPT-4.1 Miniwon 9 · lost 1

Command Awon 8 · lost 1

GPT OSS 120Bwon 8 · lost 1

G9v3 39A5Bwon 10 · lost 2

Granite 4.2 8Bwon 9 · lost 2

K2 Horizon 3.7Bwon 9 · lost 2

Grok 4.3won 8 · lost 2

Gemini 3.5 Flash Litewon 8 · lost 2

GPT-5.4 Miniwon 8 · lost 2

GPT-5.4 Nanowon 8 · lost 2

Qwen3.5 9Bwon 8 · lost 2

Llama 4 Maverick Instructwon 8 · lost 2

K2 Horizon 0.9Bwon 7 · lost 2

Gemma 4 12Bwon 7 · lost 2

K2 Horizon 7Bwon 8 · lost 3

Qwen3.6 Pluswon 7 · lost 3

Kimi K2.6won 7 · lost 3

DeepSeek V3won 7 · lost 3

Inklingwon 9 · lost 5

Gemma 4 31Bwon 9 · lost 5

GLM-4.6won 6 · lost 4

Gemini 3.6 Flashwon 6 · lost 4

Qwen3.7 Maxwon 6 · lost 4

Muse Glimmer-30Bwon 7 · lost 5

MiniMax M3won 8 · lost 6

GPT-4.1won 5 · lost 4

Mimo V2.5 Prowon 5 · lost 4

Gemini 3.1 Prowon 5 · lost 5

Nemotron 3 Ultrawon 7 · lost 7

Gemini 3.1 Flash Litewon 4 · lost 5

Qwen3 235B A22B Instructwon 4 · lost 5

Kimi K2.5won 4 · lost 5

Claude Sonnet 4.6won 4 · lost 6

DeepSeek V3.2won 4 · lost 6

Gemini 3.5 Flashwon 4 · lost 6

O4 Miniwon 4 · lost 6

Gemini 2.5 Prowon 4 · lost 6

GPT-5 Miniwon 4 · lost 6

Claude Sonnet 4won 4 · lost 6

Kimi K2won 3 · lost 6

DeepSeek R1won 3 · lost 6

Muse Spark 1.1won 3 · lost 7

Claude Sonnet 4.5won 3 · lost 7

GPT-5.2won 3 · lost 7

GPT-5.4won 3 · lost 7

Grok 4.5won 3 · lost 7

DeepSeek V4 Flashwon 3 · lost 8

GPT-5won 2 · lost 7

Gemini 3 Prowon 2 · lost 7

Qwen3.8 27Bwon 3 · lost 11

GLM-5.3 Flashwon 2 · lost 8

Claude Opus 4.5won 2 · lost 8

DeepSeek V4 Prowon 2 · lost 9

GPT-5.6 Lunawon 2 · lost 12

Muse Spark 1.2won 1 · lost 8

Grok 4won 1 · lost 8

Gemini 2.5 Flashwon 1 · lost 9

O3won 1 · lost 9

Gemini 3.8 Flashwon 1 · lost 9

Grok 4.6won 1 · lost 9

Qwen3.8 Maxwon 1 · lost 9

DeepSeek V4.1 Flashwon 1 · lost 10

GPT-5.6 Solwon 1 · lost 10

GLM-5.2won 1 · lost 13

Claude Sonnet 5won 1 · lost 13

K2 Horizon 375B A23Bwon 1 · lost 13

Claude Opus 4.8won 0 · lost 10

Claude Opus 4.7won 0 · lost 10

Gemini 3.7 Flashwon 0 · lost 10

GPT-5.5won 0 · lost 10

Claude Fable 5.1won 0 · lost 10

Claude Fable 5won 0 · lost 10

GLM-5.3won 0 · lost 11

Claude Opus 5won 0 · lost 11

Kimi K3won 0 · lost 11

GPT-5.6 Terrawon 0 · lost 14

§ 3 · Sources

Where the numbers come from

4 publications, 12 figures. Every one links to the page it was read from.

Artificial Analysis ↗1 measureread 2026-09-26

Artificial Analysis Intelligence Index 25

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 1181

AA-Briefcase ↗1 measureread 2026-09-26

AA-Briefcase 1168

✱✱ IFM — K2 Horizon launch ↗the model’s own publisher9 measuresread 2026-09-26

tau3-Banking 26.8%Terminal-Bench 2.1 58.6%SciCode 38.9%Humanity's Last Exam (without tools) 25.2%GPQA Diamond 80.8%CritPt 2.1%AA-LCR 66.3%AA-Omniscience Accuracy 18.8%AA-Omniscience Non-Hallucination 69.2%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=k2-horizon-mova-36b-a4b)](https://publicai.io/model-index/m/k2-horizon-mova-36b-a4b)