Docs
Sign inJoin the waitlist

Limits

Limits keep one workspace’s burst from slowing everyone else, and cap what a mistake, such as a script stuck in a loop, can cost. They count per workspace, across all its members, keys and agents.

Rate limits and caps

Adding a card moves a workspace to Tier 1, and Team starts at Tier 2. It moves up a tier on its own once it has paid the amount in the first row.

Limit Free, no card Tier 1 Tier 2 Tier 3 Tier 4
Paid to date $0 $50 $250 $1,000
Calls to a Model, a minute 60 600 1,200 3,000 6,000
Tokens to a Model, in and out, a minute 200,000 2,000,000 4,000,000 10,000,000 20,000,000
Calls to a Model in flight at once 8 32 64 64 64
API and agent reads, a minute 300 1,200 2,400 6,000 12,000
API and agent writes, a minute 60 240 480 1,200 2,400
Drafts, Changes and other costly starts, a minute 10 30 60 150 300
Changes running at once 1 3 5 10 20
Drafts being written at once 2 5 10 25 50
Highest spend limit $100 $250 $1,000 $5,000 $20,000

Past a per minute limit, calls answer 429 with a Retry-After header; the OpenAI SDKs wait and retry on their own. Past a cap on running work, a new change waits in the queue and starts when one finishes, and a new draft is refused with draft_busy. Drafts started a month: 10 without a card, 100 with one. Need more? Write to support@halfspacelabs.com.

Spend limit

Each workspace has one monthly spend limit, set by an Owner in Settings, Billing, up to its tier’s highest spend limit above. At 100%, new changes and calls stop until an Owner raises it or the month ends. You see every price before anything runs, and a change never spends more than its quote.

Size limits

What Limit
Rows sent with a new Model’s description 500
Rows added to an eval at once 1,000
Rows in one upload (create_upload) 20,000 training, or 5,000 eval
Longest input plus answer, in tokens 32,768
Longest input of a Model that reads long inputs in windows 64 windows of 4,000 characters; a longer one is refused with 413 input_too_long
Largest request body of one call 1 MiB; a larger one is refused with 413 request_too_large
Longest answer Set per Model; a max_tokens above it is refused with a message that names it
One call’s answers 1 (n above 1 is refused)

Next steps

  • Errors: rate_limit_exceeded, rate_limited and draft_busy
  • Pricing: plans and the spend limit
Esc