vs
5:5
overall capability wins (0 ties across 10 scored). Hy3 leads overall.
- Reasoning4:1
- Coding1:1
- Tool use0:1
- Composite indices0:2
- Specs0:2
Verdict
Hy3 edges reasoning & knowledge; K2 Horizon 375B A23B offers ~2.0x larger context (524K vs 256K).
- Hy3 edges reasoning & knowledge
- K2 Horizon 375B A23B offers ~2.0x larger context (524K vs 256K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Hy3
- $0.26
- K2 Horizon 375B A23B
- -
Price per 1M tokens
Full comparison
Organization
Tencent
MBZUAI Institute of Foundation Models
Family
Other
Other
License
Apache 2.0
Apache 2.0
Open weights
Yes
Yes
Release
Jul 6, 2026
Sep 3, 2026
Knowledge cutoff
-
-
API / provider
Tencent Cloud TokenHub
No primary API price listed
Modalities
text → text
text → text
Specs
Context window
256K
524K
Max output
-
-
Parameters
295B (21B act.)
375B (23B act.)
Pricing
Input $/1M
$0.13
—
Output $/1M
$0.53
—
Blended $/1M (3∶1)
$0.23
-
Speed
tok/s
-
-
TTFT (s)
-
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
89.7
87.3
Humanity's Last Exam
33.5
32.0
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
32.0
18.2
AA-LCR v1.1
79.0
80.0
CritPt
4.9
4.6
MMMU-Pro
—
—
IFBench
—
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
64.4
71.9
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
48.6
42.9
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
1,513
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,456
—
| Metric | Hy3 | K2 Horizon 375B A23B | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Tencent | MBZUAI Institute of Foundation Models | — |
| Family | Other | Other | — |
| License | Apache 2.0 | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Jul 6, 2026 | Sep 3, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Tencent Cloud TokenHub | No primary API price listed | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 256K | 524K | B +268K |
| Max output | - | - | — |
| Parameters | 295B (21B act.) | 375B (23B act.) | B +80 |
| Pricing | |||
| Input $/1M | $0.13 | — | — |
| Output $/1M | $0.53 | — | — |
| Blended $/1M (3∶1) | $0.23 | - | — |
| Speed | |||
| tok/s | - | - | — |
| TTFT (s) | - | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 89.7 | 87.3 | A +2.4 pts |
| Humanity's Last Exam | 33.5 | 32.0 | A +1.5 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 32.0 | 18.2 | A +13.8 pts |
| AA-LCR v1.1 | 79.0 | 80.0 | B +1.0 pts |
| CritPt | 4.9 | 4.6 | A +0.3 pts |
| MMMU-Pro | — | — | — |
| IFBench | — | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 64.4 | 71.9 | B +7.5 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 48.6 | 42.9 | A +5.7 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | 1,513 | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,456 | — | — |
Hy3
- Context
- 256K
- Parameters
- 295B (21B act.)
- Price
- $0.13 / $0.53
- Speed
- —
- Modalities
- text → text
- License
- Apache 2.0
Tencent's July 2026 general-availability Hunyuan 3 release (295B/21B-active MoE, Apache 2.0) succeeding the April 2026 Hy3 preview with expanded post-training on feedback from 50+ products.
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.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.