vs
7:0
overall capability wins (0 ties across 7 scored). K2 Horizon 375B A23B leads overall.
- Reasoning4:0
- Coding1:0
- Specs2:0
Verdict
K2 Horizon 375B A23B leads coding (widest gap: +56.9 on TermBench); K2 Horizon 375B A23B offers ~4.0x larger context (524K vs 131K).
- K2 Horizon 375B A23B leads coding (widest gap: +56.9 on TermBench)
- K2 Horizon 375B A23B offers ~4.0x larger context (524K vs 131K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- K2 Horizon 375B A23B
- -
- K2 Think V2
- -
Price per 1M tokens
Full comparison
Organization
MBZUAI Institute of Foundation Models
MBZUAI / G42 / LLM360
Family
Other
Other
License
Apache 2.0
Apache 2.0
Open weights
Yes
Yes
Release
Sep 3, 2026
Jan 27, 2026
Knowledge cutoff
-
-
API / provider
No primary API price listed
No primary API price listed
Modalities
text → text
text → text
Specs
Context window
524K
131K
Max output
-
-
Parameters
375B (23B act.)
70B
Pricing
Input $/1M
—
—
Output $/1M
—
—
Blended $/1M (3∶1)
-
-
Speed
tok/s
-
-
TTFT (s)
-
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
87.3
71.3
Humanity's Last Exam
32.0
10.1
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
18.2
18.0
AA-LCR v1.1
80.0
57.0
CritPt
4.6
—
MMMU-Pro
—
—
IFBench
—
62.8
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
71.9
15.0
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
42.9
—
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
—
| Metric | K2 Horizon 375B A23B | K2 Think V2 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | MBZUAI Institute of Foundation Models | MBZUAI / G42 / LLM360 | — |
| Family | Other | Other | — |
| License | Apache 2.0 | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Sep 3, 2026 | Jan 27, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | No primary API price listed | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 524K | 131K | A +393K |
| Max output | - | - | — |
| Parameters | 375B (23B act.) | 70B | A +305 |
| Pricing | |||
| Input $/1M | — | — | — |
| Output $/1M | — | — | — |
| Blended $/1M (3∶1) | - | - | — |
| Speed | |||
| tok/s | - | - | — |
| TTFT (s) | - | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 87.3 | 71.3 | A +16.0 pts |
| Humanity's Last Exam | 32.0 | 10.1 | A +21.9 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 18.2 | 18.0 | A +0.2 pts |
| AA-LCR v1.1 | 80.0 | 57.0 | A +23.0 pts |
| CritPt | 4.6 | — | — |
| MMMU-Pro | — | — | — |
| IFBench | — | 62.8 | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 71.9 | 15.0 | A +56.9 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 42.9 | — | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | — | — |
K2 Horizon 375B A23B
- Context
- 524K
- Parameters
- 375B (23B act.)
- Price
- —
- Speed
- —
- Modalities
- text → text
- License
- Apache 2.0
Flagship model (375B MoE / 23B active) in the Institute of Foundation Models' K2 Horizon fleet, released fully open — weights, code, training data, and methodology — under Apache 2.0.
K2 Think V2
- Context
- 131K
- Parameters
- 70B
- Price
- —
- Speed
- —
- Modalities
- text → text
- License
- Apache 2.0
MBZUAI's Institute of Foundation Models, with G42 and LLM360, fully open 70B dense reasoning model built end-to-end as a 'sovereign' successor to the original K2-Think; no hosted API yet.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.