Advanced 49B parameter reasoning model based on Llama 3.3 architecture, optimized through Neural Architecture Search with reasoning ON/OFF modes. Trained using knowledge distillation with synthetic data from advanced models. Supports 128K context window with improved efficiency over v1.
Added Aug 8, 2025
Context Window
128.0K
Max Output
16.4K
Input Price (Auto)
$0.050/1M
Output Price (Auto)
$0.25/1M
Performance metrics and benchmarks
Sourced from Artificial Analysis.
Intelligence Index
14.6
Auto routing is available for this model. Explicit provider selection is not available.
Loading provider options…
Coding Index
10.5
Agentic Index
8.4
GPQA Diamond
Graduate-level scientific reasoning
48.1%
Better than 31% of models compared
HLE
Humanity's Last Exam
4.3%
Better than 23% of models compared
IFBench
Instruction-following benchmark
32.9%
Better than 26% of models compared
T²-Bench Telecom
Conversational AI agents in dual-control scenarios
25.1%
Better than 34% of models compared
AA-LCR
Long context reasoning evaluation
22.0%
Better than 40% of models compared
GDPval-AA
Economically valuable tasks
0.0%
Better than 18% of models compared
CritPt
Research-level physics reasoning
0.0%
Better than 36% of models compared
SciCode
Python programming for scientific computing
23.8%
Better than 34% of models compared
Terminal-Bench Hard
Agentic coding and terminal use
3.8%
Better than 28% of models compared
AIME 2025
American Invitational Mathematics Examination 2025
8.0%
Better than 13% of models compared
AIME
American Invitational Mathematics Examination
13.7%
Better than 40% of models compared
MMLU-Pro
Professional and academic subject knowledge
69.2%
Better than 34% of models compared
AA-Omniscience Accuracy
Proportion of correctly answered questions
12.2%
Better than 21% of models compared
Last updated May 15, 2026
Artificial AnalysisLiveCodeBench
Contamination-free coding benchmark
29.0%
Better than 33% of models compared
Math-500
Diverse mathematical problem solving benchmark
77.0%
Better than 40% of models compared
AA-Omniscience Hallucination Rate
Rate of incorrect answers among non-correct responses
65.9%
Better than 88% of models compared
