Octomind Hub

Models for octomind.
Zero keys.

Skip collecting API keys from five providers. One login — every model works, in the CLI and in the cloud. Free models forever; the good ones in the plan. Open to everyone today — no invite, no waitlist.

# from zero to a thinking agent
$ curl -fsSL https://octomind.run/install.sh | bash
$ octomind login # approve in the browser — no keys to paste
$ octomind # models just work — pick any with -m octohub:…

Transparent prices

Published per-token prices, metering that matches the table exactly. Caps you can see live — never a surprise bill.

Free tier, forever

A capable free model with a daily allowance — even when caps or credits run out, your agent degrades gracefully instead of dying.

One key for models everywhere

The same hub key serves model calls in the CLI, inside cloud machines, and from any OpenAI-compatible tool pointed at the endpoint. (Scripting machines and sessions is the separate Developer API, documented in the panel.)

The table

11 open coding models included in the plan · 12 premium models via prepaid credits · 10 embedding models. Same table the panel and the meter use.

ModelInput /1MOutput /1MContext
Included in every paid plan
DeepSeek V4 Flash free tier$0.14$0.281M
DeepSeek V4 Pro $0.43$0.871M
Qwen3.7 Plus $0.32$1.28256k
Qwen3.7 Max $1.25$3.75256k
Gemma 4 31B $0.20$0.20128k
MiniMax M3 $0.60$2.401M
Kimi 2.6 $0.60$2.50256k
Kimi 2.7 Code $0.95$4.00256k
GLM-5.2 $1.40$4.40200k
Inkling $1.87$4.681M
Kimi K3 $3.00$15.001M

Pro caps: $20 per 4-hour window · $60/week · $120/month of usage for $20/mo. Prices track the octomind version we run.

Embeddings route through the hub too — Voyage, Jina and OpenAI models from $0.02/1M tokens, billed from prepaid credits on any plan. Local CPU embeddings stay free on every plan, hub or not.

Fair use is per account, not per key: 1 model call at a time on Free, 2 in parallel on Pro, 5 on Max, 15 pooled on Team — past that, requests queue briefly and then 429.

Prefer to self-host?

The gateway is octohub — open source, Rust, one binary. Run your own with your own provider keys; the hosted hub is the same code with the keys problem solved.

octohub (open source) →

One subscription. Models and machines.

Hub isn't a separate bill — the same plan powers hosted models here and full cloud machines. Use it straight from the panel, or drop the key into your own CLI as a provider. $10 your first month, then $20/mo. Free tier stays free.