Key Specifications
| Vendor | other |
| Version | sonar-reasoning |
| Release Date | 2024-12-18 |
| Context Window | 127072 tokens |
| Input Modalities | text |
| Output Modalities | text |
| License | Proprietary |
| Documentation | https://huggingface.co/models |
Benchmark Performance
| Benchmark |
Score |
Unit |
Evaluated At |
Notes |
Source |
| MMLU |
87.7 |
% |
2024-12-18 |
5-shot |
view |
| HUMANEVAL |
81.1 |
pass@1 |
2024-12-18 |
— |
view |
| GSM8K |
89.8 |
% |
2024-12-18 |
0-shot CoT |
view |
| MATH |
50.7 |
% |
2024-12-18 |
0-shot CoT |
view |
| BBH |
83.4 |
% |
2024-12-18 |
3-shot CoT |
view |
| GPQA |
50.1 |
% |
2024-12-18 |
0-shot |
view |
| IFEVAL |
83.1 |
% |
2024-12-18 |
prompt_strict |
view |
| ARC |
96.2 |
% |
2024-12-18 |
challenge |
view |
| MUSR |
69.4 |
% |
2024-12-18 |
0-shot |
view |
| WINOGRANDE |
87.2 |
% |
2024-12-18 |
0-shot |
view |
Pricing
| Tier |
Price |
Currency |
| Input | $2 / Mtok | USD |
| Output | $8 / Mtok | USD |
| Cache Read | $0 / Mtok | USD |
| Cache Write | $0 / Mtok | USD |
Source:
https://huggingface.co/models
· as of 2024-12-18
Compliance
- Data Residency: self-host
- SOC2: ✗
- HIPAA: ✗
- GDPR: ✗
- ISO 27001: ✗
Sonar Reasoning
Model Overview
Perplexity Sonar Reasoning 推理型在线 RAG 模型, 127K 上下文, 基于 DeepSeek R1 微调, 链式思维推理。
Core Specifications
| Vendor |
Version |
Release Date |
Context Window |
Input Modalities |
Output Modalities |
License |
| Other |
sonar-reasoning |
2024-12-18 |
127K |
text |
text |
Proprietary |
| Benchmark |
Score |
Unit |
Notes |
| MMLU (Massive Multitask Language Understanding) |
87.7 |
% |
5-shot |
| HumanEval |
81.1 |
pass@1 |
— |
| GSM8K (Grade School Math 8K) |
89.8 |
% |
0-shot CoT |
| MATH |
50.7 |
% |
0-shot CoT |
| BBH (BIG-Bench Hard) |
83.4 |
% |
3-shot CoT |
| GPQA |
50.1 |
% |
0-shot |
| IFEval |
83.1 |
% |
prompt_strict |
| ARC |
96.2 |
% |
challenge |
| MUSR |
69.4 |
% |
0-shot |
| WinoGrande |
87.2 |
% |
0-shot |
Pricing
| Input |
Output |
Cache Read |
Cache Write |
| — |
— |
— |
— |
per million tokens
Strengths
- MMLU score 87.7, strong knowledge reasoning.
- HumanEval 81.1, excellent code generation.
- GSM8K 89.8, robust math reasoning.
Weaknesses
- Proprietary, not self-hostable.
Use Cases
- Code generation and debugging
References