The LLM benchmark aggregator.

K2 Horizon 375B A23B

Institute of Foundation Models · 375B · A23B

Strongest in Knowledge work (#31 of 178), weakest in General intelligence (#72 of 204). Above par in 17 of 17 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5 and Claude Fable 5 and ahead of MiniMax M3 and Gemini 3.5 Flash.

§ 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.

Agents59.3−8.8#37/2682/5

Knowledge work62.1−11.5#31/1782/2

Tool use60.1−13.9—/81✱0/1

Web research50.5−13.7—/4✱0/1

Workflow automation50.8−11.2—/0✱✱

Core abilities52.3−15.1#69/2041/3

General intelligence53.3−16.8#72/2041/3

Long context59.3−5.3—/0✱✱

Reasoning51.3−15.6—/178✱0/4

Science54.3−9.2—/122✱0/1

Expert reasoning52.4−8.5—/0✱✱

Coding52.5−16.6—/165✱0/5

Agentic coding51.8−16.9—/157✱0/4

Code generation53.8−16.2—/77✱0/2

Repository Q&A60.7−6—/0✱✱

Knowledge52.3−11.2—/138✱0/2

Factuality53.4−5.6—/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 375B A23B, left for the other.

Granite 4.2 3Bwon 11 · lost 0

Granite 4.2 8Bwon 11 · lost 0

G9v3 39A5Bwon 12 · lost 0

K2 Horizon 32Bwon 12 · lost 0

Granite 4.2 30Bwon 13 · lost 0

Nemotron 3 Superwon 13 · lost 0

Qwen3.6 35B A3Bwon 14 · lost 0

K2 Horizon MoVA 36B A4Bwon 13 · lost 1

K2 Horizon 3.7Bwon 10 · lost 1

Nemotron 3 Ultrawon 14 · lost 2

K2 Horizon 7Bwon 10 · lost 2

Gemma 4 31Bwon 11 · lost 3

Inklingwon 13 · lost 4

Muse Glimmer-30Bwon 9 · lost 3

Gemini 3.5 Flashwon 8 · lost 3

MiniMax M3won 10 · lost 7

DeepSeek V4 Flashwon 6 · lost 7

GLM-5.2won 7 · lost 9

Grok 4.5won 4 · lost 7

DeepSeek V4 Prowon 4 · lost 9

GPT-5.6 Terrawon 4 · lost 11

GPT-5.6 Lunawon 4 · lost 12

DeepSeek V4.1 Flashwon 3 · lost 10

Kimi K3won 3 · lost 10

Qwen3.8 27Bwon 3 · lost 11

GPT-5.6 Solwon 3 · lost 11

GLM-5.3won 2 · lost 11

Claude Sonnet 5won 2 · lost 14

Claude Fable 5won 0 · lost 11

Claude Opus 5won 0 · lost 13

§ 3 · Sources

Where the numbers come from

5 publications, 27 figures. Every one links to the page it was read from.

Artificial Analysis ↗1 measureread 2026-09-26

Artificial Analysis Intelligence Index 31

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 1349

AA-Briefcase ↗1 measureread 2026-09-26

AA-Briefcase 1295

✱✱ IFM — K2 Horizon launch ↗the model’s own publisher18 measuresread 2026-09-26

GDPVal-AA 1441tau3-Banking 34%Toolathlon Verified 65.3%Automation Bench Public 25.3%Apex-Agents (pass@1) 24.8%MCPMark 67.7%BrowseComp 72.8%WildClawBench 50.9%Terminal-Bench 2.1 70.2%SciCode 42.7%SWE-Atlas-QnA 48.4%SWE Bench Pro 42.6%Humanity's Last Exam (without tools) 32%GPQA Diamond 87.3%CritPt 8.6%AA-LCR 76%AA-Omniscience Accuracy 23%AA-Omniscience Non-Hallucination 74.7%

✱ CellCog — K2 Horizon analysis ↗6 measuresread 2026-09-26

GDPVal-AA (Elo) 1441Toolathlon Verified 65.3%Terminal-Bench 2.1 70.2%SWE-bench Pro (strict) 42.6%MCPMark 67.7%SWE-Atlas-QnA (strict) 48.4%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=k2-horizon-375b-a23b)](https://publicai.io/model-index/m/k2-horizon-375b-a23b)