Blackbox

What it is for

Blackbox is a hosted encrypted inference and agent platform: one endpoint for a 300+ model catalog, plus a cloud IDE agent that clones a repo and writes code. Use it when the job is tokens that stay sealed in the client, not when you need a $20 IDE seat.

It is not open source. Public pricing is enterprise commit + per-1M-token rates. There is no self-serve $8 Pro card on the page we fetched.

What makes it unique: the cover is Blackbox’s official cloud-IDE still — agent.blackbox.ai workspace, a Dropbox clone task, BLACKBOXAI writing an implementation plan next to the editor.

Key features

• End-to-end encrypted inference — the key is sealed in the client; the proxy, host, and operator never hold it (blackbox.ai).
• One endpoint, one bill — route the catalog through a single API / CLI instead of a provider key per model (Pricing).
• Cloud coding agents — Agents API plus the hosted IDE agent that clones a repo, browses, and edits (cover; Pricing).
• GPU rental with controls — scoped API keys, spend caps, SSH fingerprint checks, and one audit log; your team never holds a cloud-provider credential (blackbox.ai).
• Bring your own provider accounts — connect OpenAI / Anthropic / Google and keep their negotiated rates while Blackbox handles routing, failover, and cache (Pricing).
• Forward-deployed engineer + agent loop — an FDE stands up the endpoint; agents propose engine changes that a fixed harness keeps only if faster (blackbox.ai).

Who it's for

Best fit: security-sensitive teams that will sign an annual token commit and want encryption on every hop.

Poor fit: individual developers who want a monthly IDE seat (Cursor or GitHub Copilot).

Pricing

Verified 2026-10-22 (Asia/Shanghai) from blackbox.ai/pricing. There is no platform fee and no per-seat charge. You talk to sales, commit annual spend, and burn it at (or below) the published per-1M-token rates. Unused commit expires at period end (alerts at 75% and 90%).

Published list rates (input / output / cache-read per 1M tokens), used as the contract reference:

• DeepSeek V4.1 Flash — $0.07 / $0.30 / $0.002
• Kimi K3 — $3.00 / $15.00 / $0.30
• GLM-5.3 — $1.40 / $4.40 / $0.14
• Qwen3.8 Max — $2.00 / $6.00 / $0.25

Enterprise includes a dedicated FDE, implementation at $0, single-tenant deploy, data residency, SAML/SCIM/RBAC, zero retention + DPA, and custom TPM. Commit discounts are quoted against those list rates (page copy: Closed −5%+ / Open −10%+).

Do not invent a leftover consumer “Pro $8/mo” SKU. Confirm the PO.

Limitations / costs

• Sales-gated from the start. There is no public self-serve checkout on the pricing page we fetched.
• Unused commit expires. Size it to minimum usage.
• This is not a $20 IDE. The cloud agent is metered in tokens, not a Copilot-style seat.
• Bring-your-own-key still bills the provider. Routing does not erase that invoice.

Alternatives

• Groq — speed-first inference if you do not need client-sealed encryption.
• Fireworks AI — hosted inference with a public token card, no annual PO required to start.
• Together AI — open-weight inference if the job is a model API, not an encrypted agent IDE.

Official tutorial

Official BLACKBOX AI channel video: BLACKBOXAI Agent on a cloud based IDE. No Downloads Required. (oEmbed author: BLACKBOX AI).

Written path: blackbox.ai plus Pricing.

Website

https://www.blackbox.ai

Source

Catalog row curated for AI Tools Nav. Cover is Blackbox’s official cloud-IDE agent still. Enterprise commit + token rates (Flash $0.07/$0.30 per 1M). Tutorial: cloud IDE agent on @blackboxbruiser (oEmbed).

Last verified

2026-10-22 (Asia/Shanghai) editorial check.