Compare models
Up to four models side by side. The best value in each row is highlighted. Task scores are heuristics, not benchmark results.
| Attribute | |||
|---|---|---|---|
| Overview | |||
| Company | DeepSeek | Alibaba (Qwen) | OpenAI |
| Version | 4 | 3.5 | — |
| Tier | frontier | balanced | fast |
| Released | 2026-08-12 | 2026-02-25 | 2025-08-05 |
| Open weights | Yes | Yes | Yes |
| Parameters | 1.65T | 125.1B (10B active) | 117B (5.1B active) |
| Status | Active | Active | Active |
| Pricing (USD per 1M tokens) | |||
| Input | $0.12 | $0.26 | $0.014 (best in this row) |
| Output | $4.60 | $2.08 | $0.052 (best in this row) |
| Cached input | $0.11 | — | $0.013 (best in this row) |
| Batch input | — | — | — |
| Batch output | — | — | — |
| Blended (3 in : 1 out) | $1.24 | $0.72 | $0.024 (best in this row) |
| Est. cost per token | $0.000001236 | $0.000000715 | $0.000000023 (best in this row) |
| Capacity | |||
| Context window | 1.05M (best in this row) | 262K | 131K |
| Max output | 944K (best in this row) | 66K | 66K |
| Inputs | text | image, text, video | text |
| Features | |||
| Caching | Yes | No | Yes |
| JSON schema | Yes | Yes | Yes |
| Reasoning | Yes | Yes | Yes |
| Tools | Yes | Yes | Yes |
| Task fit (heuristic, 1–5) | |||
| coding | 5 / 5 (best in this row) | 4 / 5 | 3 / 5 |
| reasoning | 5 / 5 (best in this row) | 4 / 5 | 3 / 5 |
| math | 5 / 5 (best in this row) | 4 / 5 | 3 / 5 |
| summarization | 5 / 5 (best in this row) | 4 / 5 | 4 / 5 |
| extraction | 5 / 5 (best in this row) | 4 / 5 | 4 / 5 |
| classification | 5 / 5 (best in this row) | 4 / 5 | 4 / 5 |
| chat | 5 / 5 (best in this row) | 4 / 5 | 3 / 5 |
| creative | 5 / 5 (best in this row) | 4 / 5 | 3 / 5 |
| translation | 5 / 5 (best in this row) | 4 / 5 | 4 / 5 |
| vision | 1 / 5 | 4 / 5 (best in this row) | 1 / 5 |
| long context | 5 / 5 (best in this row) | 3 / 5 | 3 / 5 |
| agentic | 5 / 5 (best in this row) | 4 / 5 | 3 / 5 |
Prices are USD per 1M tokens from vendor price lists. “—” means the vendor does not publish it.