engyPricing
Per-token, pay as you go. No subscriptions and no minimums; you pay only for the tokens you use.
| model | input↑ | output | cached | vs official | ctx |
|---|---|---|---|---|---|
MiniMax-H3new | $0.03/second | — | 32K | ||
deepseek-v4.1-flashnewZDR | $0.30$0.04 | $1.20$0.08 | $0.006$0.008 | −90% | 328K |
deepseek-v4-flash-0731ZDR | $0.045 | $0.09 | $0.009 | — | 1M |
qwen3.6-35b-a3b | $0.375$0.045 | $2.25$0.30 | $0.015 | −87% | 208K |
qwen3.8-27bZDR | $0.50$0.045 | $3.00$0.32 | $0.015 | −90% | 1M |
glm-5.3-flash(Ox Alpha)newZDR | $0.15$0.135 | $0.50$0.45 | $0.03$0.027 | −10% | 262K |
glm-5.2ZDR | $1.40$0.68 | $4.40$1.50 | $0.26$0.18 | −59% | 262K |
glm-5.3newZDR | $1.40$0.98 | $4.40$3.08 | $0.26$0.18 | −30% | 328K |
kimi-k3ZDR | $3.00$1.95 | $15.00$9.75 | $0.30$0.195 | −35% | 1M |
| $ per 1M tokens (/second = $ per second of output) · $1.40 = the model developer's own API price, checked 2026-09-29 · vs official uses a 3:1 input:output blend | |||||
Prompt-cache hits bill at the cached rate automatically, with no config and no cache_control markers. Agentic workloads (coding assistants, multi-turn tools) typically hit 90%+ cache on repeated prefixes, so effective input cost is usually far below the headline rate.
ZDR marks models under zero data retention: served only on hardware engy operates, prompts and outputs never stored or trained on. MiniMax-H3 (not covered by engy's zero-data-retention configuration) and qwen3.6-35b-a3b (routed across permissionless subnet miners, on hardware without confidential computing) carry no such mark; see terms and privacy.
Prices are live from the billing engine. The API reports the same numbers at https://api.engy.ai/v1/models.