Limits

Free answers

Six models answer free on POST /v1/chat/completions, with no wallet or key: free/openai-gpt-oss-20b, free/gemma-4-e2b, free/gemma-3-27b-it, free/mistral-small-3.2-24b-instruct, free/qwen3-coder-30b-a3b-instruct and free/deepseek-v4.1-flash. The default allowance (shared by all six) is 20 answers per visitor network per day: IPv4 /24 or IPv6 /64. IPv4-mapped IPv6 addresses share their IPv4 /24 allowance. Daily counters reset at UTC midnight. Shared spend and answer counts persist across restarts; visitor counts remain in memory.

These are configurable server defaults. A shared daily budget of $0.25 caps what free answers cost upstream and can stop them earlier: once it is spent, free requests return 429 free_budget_exhausted with Retry-After until midnight UTC. Daily exhaustion returns 429 until the reset; concurrency exhaustion returns free_tier_busy with Retry-After: 5. A failed free answer is returned to the visitor allowance.

Request and lookup limits

The server JSON body limit is 128 KiB (131,072 bytes). /v1/prices allows 60 requests per visitor per minute and 3000 overall in fixed minute windows. On 429, wait Retry-After: 60. See Errors for reset headers.

API keys

Keys have server-enforced daily and per-request caps and an expiry. Documented defaults are $5 per day, $0.25 per request and 90 days, with 60 requests per minute. Documented upper bounds are $1000 per day, $50 per request, 365 days and 10 active keys per account. The key caps cannot exceed the account caps. USDC top-ups range from $1 to $50 at a time.

Payment ceilings and client budgets

The pricing module sets a minimum authorization ceiling of 1500 USDC micro-units (0.0015 USDC). This is an authorization floor, not a minimum actual charge for upto. Default TERMS.maxChargeAtomic is 3000 micro-units (0.003 USDC); request-specific quotes replace that default with a ceiling calculated for the question, model and output budget. Read the actual offer rather than assuming this default is a universal cap.

For upto, billing refuses an actual charge above the authorized maxChargeAtomic. exact settles its flat offered amount; an actual cost overrun is absorbed by the service. The TypeScript SDK options maxPerCallUsd and maxPerDayUsd are client-side guards. Its daily counter is in memory per SDK instance and UTC day; a new process or instance starts a new counter. They do not configure server or wallet limits.

Solana

Base and Solana use the same Ask URL and chat offers include both upto and exact on both networks. Model answers have a settlement fee of 0.002 USDC on Solana (0.001 on Base). This does not describe a separate Solana request quota.

For a wallet that granted SingIt an allowance, a Solana pause is lifted only by a new GRANT prepared after the pause, signed by the owner’s wallet and confirmed on chain. Changing limits alone, a failed approval or an approval prepared before the pause does not lift it. See wallet spending limits.