qwen2.5-0.5b-abliterated
Qwen2.5 0.5B Instruct, refusal direction removed (abliterated). Tiny and fast.
| Input | Cached input | Output | Minimum per request | Fastest measured | Refused, of 100 |
|---|---|---|---|---|---|
| $0.25 / 1M | $0.025 / 1M | $1.00 / 1M | $0.0002 | 1,466 tokens/s | 1 |
Models
Small models while we prove the platform. Each one lists only what has passed on real hardware.
qwen3-1.7b-abliterated
Qwen3 1.7B, refusal direction removed (abliterated). Supports thinking mode.
| Input | Cached input | Output | Minimum per request | Fastest measured | Refused, of 100 |
|---|---|---|---|---|---|
| $0.25 / 1M | $0.025 / 1M | $2.00 / 1M | $0.00025 | 1,393 tokens/s | 0 |
Models start on demand. See pricing for the start and idle fees, and benchmarks for how these were measured.