Llama-3.3-70B-Instruct vs Sonar
API pricing, benchmark indices and specs — updated 19 October 2026.
meta-llama · A · Open weights
Params
70.6 B
Context
—
30d
210 K
llama3.3 licence · conditional
perplexity · B · API-only
Input / 1M
$1.00
Output / 1M
$1.00
AA index
7.7
proprietary · 127.1K context
Specification comparison
Differences are highlighted; identical values are muted.
| Specification | Llama-3.3-70B-Instruct | Sonar |
|---|---|---|
| Deployment | Open weights · also via API | API-only (closed weights) |
| Parameters | 70.6 B | n/a — API-only |
| Architecture | LlamaForCausalLM | n/a — API-only |
| Context length | 131,072 | 127,072 |
| Licence | llama3.3 | Proprietary |
| Commercial use | Conditional | Per API terms |
| Languages | 8 | n/a — API-only |
| Downloads 30d | 209,625 | n/a — API-only |
| Downloads all time | 14.8 M | n/a — API-only |
| Quantizations on the Hub | SAFETENSORS | n/a — API-only |
| Gated | Yes — accept licence | n/a — API-only |
| First seen on the Hub | 2024-11-26 | n/a — API-only |
| API input price / 1M tokens | $0.10 | $1.00 |
| API output price / 1M tokens | $0.32 | $1.00 |
| Est. cost per 1M input + 1M output tokens | $0.42 | $2.00 |
| Cached input / 1M tokens | — | — |
| AA Intelligence Index | 7.7 | 7.7 |
| AA Coding Index | 11.9 | — |
| AA Agentic Index | — | — |
| Max output tokens | 16,384 | 114,364 |
| API providers | 9 | 1 |
| Tool calling | Yes | No |
| Reasoning mode | No | No |
API cost
Estimated cost of 1M input + 1M output tokens via OpenRouter
Llama-3.3-70B-Instruct
$0.42
Sonar
$2.00
Llama-3.3-70B-Instruct is 4.8× cheaper at list price.
VRAM side by side
On RTX 4090 · 24 GB · 8K context unless noted
Llama-3.3-70B-Instruct · fp16
166.3 GB / 24 GB