Specs
On paper
| Llama 3.3 70B Instruct Turbo | GPT-5 | |
|---|---|---|
| Modality | Chat | Chat |
| Context | 131K | 128K |
| Max output | 16K | 16K |
| Price | $0.002typical | $0.025typical |
| Metered input | $0.25/M tok | $1.96/M tok |
| Metered output | $0.47/M tok | $15.68/M tok |
| Providers | 1 | 1 |
| Streaming | ||
| JSON mode | ||
| Tool calling |
Live performance
On our traffic
Each curve is one model, measured on the same network, the same days.
Response latencymedian per day
Collecting live data — a point is recorded for every day of traffic.
Output speedtokens / second
Collecting live data — a point is recorded for every day of traffic.
Uptimesuccessful of gradable
Collecting live data — a point is recorded for every day of traffic.
Requests / dayvolume served
Collecting live data — a point is recorded for every day of traffic.
Llama 3.3 70B Instruct TurboGPT-5
Run both. Route to the winner.
One key serves every model here — swap the model string, keep your code.
Get an API key