Tools
Compare models
Add up to four models via ?models=slug-a,slug-b — or pick from any model page.
| Spec | DeepSeek | OpenAI |
|---|---|---|
| Params | 291B / 37B | 120B |
| Architecture | moe | moe |
| Context | 1M | 131K |
| Weights | ||
| License | MIT License | Apache License 2.0 |
| 183.4 GBest | 75.9 GBest | |
| Cheapest hosted (out/1M) | $0.28 · Novita AI | $0.17 · OpenRouter |
| Released | Apr 22, 2026 | Aug 4, 2025 |
| LiveBench | 65.5 | — |
| LiveBench Agentic Coding | 37.6 | — |
| LiveBench Coding | 69.2 | — |
| LiveBench Data Analysis | 68 | — |
| LiveBench Instruction Following | 63.1 | — |
| LiveBench Language | 70.1 | — |
| LiveBench Mathematics | 79.7 | — |
| LiveBench Reasoning | 70.6 | — |
| Arena Agent (IPS) | −0.03 | — |
| Arena Coding | 1483.5 | 1390.1 |
| Arena Creative Writing | 1407.4 | 1277.5 |
| Arena Hard Prompts | 1459.3 | 1362 |
| Arena Instruction Following | 1428.3 | 1325.3 |
| Arena Maths | 1426.6 | 1381.3 |
| Arena Text (overall) | 1435.6 | 1352.3 |
| Arena Code (WebDev) | 1576.5via High | — |
| SWE-bench Verified | — | 26via mini-SWE-agent |
