Rate card

Here's exactly what everything costs.

Platform meters

Documents, workflows, and compute

Document intelligence

Scanned documents1.02 cr / page
Digital documents0.02 cr / page

The page price covers everything — reading the document, understanding it, and making it searchable. No separate OCR or indexing fees.

Workflow runs

1 cr / run

The first 25 steps of every run are included; each step beyond that adds 0.1 credits.

Sandbox compute

0.1 cr / minute

Agent sandbox executions are metered per wall-clock minute, rounded up.

Model rates

Every model, credits per million tokens

Reading cached context costs ~80% less than fresh input, so agents that reuse long context save significantly on every run. Models marked "Included in Free" are the ones the Free plan can use. A dash means the provider doesn't bill that dimension.

OpenAI credit rates per million tokens
OpenAIInputcr / 1M tokensOutputcr / 1M tokensCache readcr / 1M tokens
gpt-5.5-2026-04-23premium
1,0006,000200
gpt-5.4-2026-03-05standard
5003,000100
gpt-5.4-mini-2026-03-17standard
15090030
gpt-5.4-nano-2026-03-17Included in Free
4025010
Anthropic credit rates per million tokens
AnthropicInputcr / 1M tokensOutputcr / 1M tokensCache readcr / 1M tokensCache writecr / 1M tokens
claude-opus-4-8premium
1,0005,0002001,250
claude-sonnet-5standard
6003,000120750
claude-haiku-4-5Included in Free
2001,00040250
Google credit rates per million tokens
GoogleInputcr / 1M tokensOutputcr / 1M tokensCache readcr / 1M tokens
gemini-3.1-pro-previewpremium
4002,40080
↳ context above 200k tokens8003,600
gemini-3.5-flashstandard
3001,80060
gemini-3-flash-previewstandard
10060020
gemini-3.1-flash-liteIncluded in Free
5030010

The fine print, in plain language

  • Model rates track provider list prices. When a provider moves their prices, these rates move with them.
  • Rate changes are forward-only: usage already metered keeps the rate it was metered at. Grandfathering isn't applicable pre-launch.

The meter is public. Put it to work.

Start free with a daily credit grant, or bring one painful document workflow to a demo and watch the meter run on your actual problem.