DeepSeek V4.1 Flash and GLM 5.3 Flash, two fast open models, in any coding agent. A flat $60 a month for 1,120 credits a week.
Get a seatNext in the stable: Hy4 Flash, Tencent's 220B model, at 400+ tokens a second on our own GPU. Hold your horses: 400+ is a target until that GPU measures it.
A Canter seat gets:
1,120credits a week, shared by both models. At zero it stops; nothing more is charged.
4requests at once.
2models: DeepSeek V4.1 Flash 1017 and GLM 5.3 Flash.
| Input | Cached | Output | |
|---|---|---|---|
| DeepSeek V4.1 Flash 1017 | 13 | 0.8 | 50 |
| GLM 5.3 Flash | 5 | 3 | 20 |
A credit is one cent of what the model costs us. In our test, one Claude Code task used 2.2 credits on DeepSeek and 2.7 on GLM. Credits come back every 7 days from your billing date and do not carry over; the last few days of a billing month, short of a week, get 160 each. No daily, 5-hour or monthly caps.
Steed Code speaks both the OpenAI and the Anthropic API. It works in OpenCode, Claude Code, Cline, Kilo, Roo, OpenClaw and Hermes. In Claude Code it is three lines:
export ANTHROPIC_BASE_URL=https://api.steedlabs.cloud export ANTHROPIC_AUTH_TOKEN=your-key export ANTHROPIC_MODEL=DeepSeek-V4.1-Flash-1017
Cancel any time. A full refund within 14 days; see the refund policy.