vs
2:1
overall capability wins (0 ties across 3 scored). Atria Dawn Preview leads overall.
- Tool use1:0
- Specs1:1
Verdict
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.
- Atria Dawn Preview
- -
- K2 Horizon 375B A23B
- -
Price per 1M tokens
Full comparison
Organization
Shanghai AI Lab
MBZUAI Institute of Foundation Models
Family
Other
Other
License
MIT
Apache 2.0
Open weights
Yes
Yes
Release
Sep 11, 2026
Sep 3, 2026
Knowledge cutoff
-
-
API / provider
No primary API price listed
No primary API price listed
Modalities
text → text
text → text
Specs
Context window
256K
524K
Max output
66K
-
Parameters
744B
375B (23B act.)
Pricing
Input $/1M
—
—
Output $/1M
—
—
Blended $/1M (3∶1)
-
-
Speed
tok/s
-
-
TTFT (s)
-
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
—
87.3
Humanity's Last Exam
—
32.0
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
18.2
AA-LCR v1.1
—
80.0
CritPt
—
4.6
MMMU-Pro
—
—
IFBench
—
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
59.6
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
71.9
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 | Atria Dawn Preview | K2 Horizon 375B A23B | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Shanghai AI Lab | MBZUAI Institute of Foundation Models | — |
| Family | Other | Other | — |
| License | MIT | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Sep 11, 2026 | Sep 3, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | No primary API price listed | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 256K | 524K | B +268K |
| Max output | 66K | - | — |
| Parameters | 744B | 375B (23B act.) | A +369 |
| Pricing | |||
| Input $/1M | — | — | — |
| Output $/1M | — | — | — |
| Blended $/1M (3∶1) | - | - | — |
| Speed | |||
| tok/s | - | - | — |
| TTFT (s) | - | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | — | 87.3 | — |
| Humanity's Last Exam | — | 32.0 | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 18.2 | — |
| AA-LCR v1.1 | — | 80.0 | — |
| CritPt | — | 4.6 | — |
| MMMU-Pro | — | — | — |
| IFBench | — | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | 59.6 | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | 71.9 | — |
| 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 | — | — | — |
Atria Dawn Preview
- Context
- 256K / 66K out
- Parameters
- 744B
- Price
- —
- Speed
- —
- Modalities
- text → text
- License
- MIT
Shanghai AI Laboratory's open-weight agentic MoE (744B total, MIT license, 256K context), built on a GLM-5.2 base and trained by the cross-institutional ATRIA initiative (with Fudan NLP Lab) for long-horizon research and engineering workflows: problem analysis, tool use, code, experiments, and failure recovery. Benchmark scores are from the model's own technical report and have not yet been independently verified by a third-party evaluator.
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.