Compare models
Comparing 3models. Drop the URL into a doc — it's permalinked.
| Field | deepseek/deepseek-chat | groq/llama-3.3-70b-versatile | google/gemini-2.5-flash |
|---|---|---|---|
| Provider | deepseek | groq | |
| Model ID | deepseek-chat | llama-3.3-70b-versatile | gemini-2.5-flash |
| Context | 164K | 131K | 1.0M |
| Max output | 8K | 33K | 8K |
| Input / 1M | $0.32 | $0.59 | $0.30 |
| Output / 1M | $0.89 | $0.79 | $2.50 |
| Cached input / 1M | $0.07 | — | $0.07 |
| Avg cost / 1M | $0.60 | $0.69 | $1.40 |
| Speed | 50 t/s | 280 t/s | 250 t/s |
| Quality index | 72.0 | 56.0 | 70.0 |
| MMLU | 87.1 | 86.0 | 85.3 |
| GPQA | 59.1 | 50.5 | 65.2 |
| HumanEval | 89.0 | 88.4 | 88.1 |
| MATH | 90.2 | 77.0 | 88.0 |
| SWE-bench | — | — | — |
| Arena Elo | — | — | — |
| Tools | ✓ | ✓ | ✓ |
| Vision | — | — | ✓ |
| Thinking | — | — | ✓ |
| Streaming | ✓ | ✓ | ✓ |
| JSON mode | ✓ | ✓ | ✓ |
| Structured output | ✓ | — | ✓ |
| Prompt cache | — | — | — |
Same data, in your terminal: relay models compare deepseek-v3 llama-3.3-70b gemini-flash