Model Benchmark
Quality, latency, and price across leading models.
Quality Score — higher is better
| Model | Quality Score | Cost per 1M tokens | Latency P95 |
|---|---|---|---|
| qwen-max | 85.2 | $4/1M | 920ms |
| deepseek-v3 | 88.1 | $0.69/1M | 760ms |
| moonshot-v1-128k | 82.4 | $8.4/1M | 1080ms |
| glm-4 | 83.7 | $1.4/1M | 990ms |
| doubao-pro | 84 | $0.2/1M | 870ms |
| ernie-4.0 | 83.1 | $8.4/1M | 1010ms |
| gpt-4 | 89.6 | $30/1M | 1340ms |
| claude-3-5-sonnet | 90.1 | $9/1M | 1120ms |
| gemini-1-5-pro | 87.8 | $7/1M | 1260ms |
Last updated: 2026-05-01
Methodology
Scores are curated from third-party public sources and vendor-published figures. He-API model prices are rendered live from the public model catalogue. Quality and latency reflect the cited sources at the last-updated date.
- Vendor-published pricing pages (2026-Q2)
- Public third-party evaluation leaderboards (quality)
- Curated latency sampling, p95 over 1k requests (third-party)
Curated from third-party public sources; not a real-time automated benchmark.