List prices plus observed cost per turn, from real runs metered through cerver. Ranked by what leaves your wallet.
| Model | Input $/1M | Output $/1M | Observed $/turn | Sample | |
|---|---|---|---|---|---|
| 1 | claude-opus-4-8 Anthropic |
$15 | $75 | $0.6123 | 721 turns |
| 2 | gpt-5 OpenAI |
$5 | $15 | $0.0027 | 20 turns |
| 3 | claude-sonnet-4-6 Anthropic |
$3 | $15 | $0.0058 | 46 turns |
| 4 | grok-4 xAI |
$3 | $15 | — | no runs yet |
| 5 | gpt-4o OpenAI |
$2.5 | $10 | $0.0024 | 13 turns |
| 6 | claude-haiku-4-5 Anthropic |
$1 | $5 | $0.0033 | 384 turns |
| 7 | gpt-5-mini OpenAI |
$1 | $4 | $0.0009 | 23 turns |
| 8 | glm-5.2 Zhipu (open) |
$0.6 | $2.2 | $0.0048 | 16 turns |
| 9 | gpt-5-nano OpenAI |
$0.25 | $1 | $0.0027 | 13 turns |
Observed columns come from runs metered on cerver over the last 60 days — a young sample that grows with the platform. List prices are vendors' published rates.
The point of the spread: most of your work doesn't need the top row. On cerver, routing a task to row 6 instead of row 1 is a config change — and every run's cost lands in your dashboard, attributed per user.