ForgeAPI serves open-weight LLMs — DeepSeek, Qwen, Kimi, Llama — from our own GPU inference stack. One OpenAI-compatible API, per-token pricing, no monthly fees. Built by people who run the hardware.
No proxying to third-party clouds. Every model below runs on ForgeAPI's own inference hardware — which is why pricing stays low and stable.
Drop-in replacement. Change one line — your base_url — and existing OpenAI SDK code works with DeepSeek, Qwen, Kimi and more.
We run the GPUs. No multi-hop proxying, no shared cloud accounts — median first token under 200 ms, stable under load.
No subscriptions, no per-seat fees. Top up once and spend exactly what you use. Credits never expire.
Every account gets daily free access to smaller models — enough to build, test and prototype before you pay anything.
Requests route across a pool of upstream instances. If one node hiccups, your call retries elsewhere — invisibly.
Per-key dashboards with token spend, request volume and latency breakdowns. Know what each integration costs you.
Works with the official OpenAI SDKs — just point your base_url at ForgeAPI. Python, Node and curl examples in the docs.
# One line to switch: point base_url at ForgeAPI
curl https://api.forgeapi.dev/v1/chat/completions \
-H "Authorization: Bearer $FORGE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4-flash",
"messages": [{"role": "user", "content": "Hello!"}],
"max_tokens": 128
}'One price per model. No tiers, no enterprise sales calls. Prices below — and the full catalog is one click away.
| Model | Input / 1M tokens | Output / 1M tokens | |
|---|---|---|---|
| deepseek-v4-flash | $0.25 | $0.75 | available |
| deepseek-v4-pro | $1.00 | $3.00 | available |
| kimi-3 | $0.60 | $2.40 | available |
| qwen-3-235b | $0.40 | $1.20 | beta |
| llama-3.3-70b | $0.28 | $0.84 | available |
| qwen-3-8b (free tier) | $0 | $0 | free |
Rate-limited free tier · volume discounts available at $500+/mo · full details →
Free API key in 30 seconds. Free tier included, credit card optional.