New limits: 3,000 RPM and 1 billion tokens per day on every top-up account
Rate limits just went up 3x across the board for paying accounts: 3,000 requests per minute, 250,000 requests per day, 20M tokens per minute and 1B tokens per day. Free tier unchanged. Agent traffic rides x402 with no signup.
Two months of usage data said the same thing: developers hit our ceilings, not their own. The most common support question this month was not about quality or uptime. It was "I keep getting 429s at 50M tokens, can my account just have more".
So the limits changed. Effective immediately for every top-up account:
- Requests per minute: 3,000 (was 1,000)
- Requests per day: 250,000 (was 20,000)
- Tokens per minute: 20,000,000 (was 2,000,000)
- Tokens per day: 1,000,000,000 (was 50,000,000)
Nothing to opt into: the tier system enforces it live. The free tier (never paid) is unchanged at 30 RPM, 300 requests/day, 100,000 tokens/minute and 1,000,000 tokens/day: enough to evaluate the platform, apply for a job, or run a small script, not enough to run production traffic, which is what a top-up is for.
Why 1B tokens/day
Because the biggest verified workload on the platform is now one person: a single account processed 49.9M tokens in one session the day before the change, right against the old cap. An agentic loop re-ranking 20,000 tables, a nightly ingestion pipeline, a fleet of phones running the same assistant: none of these should need to talk to sales. At DeepSeek's pinned rates a ceiling of 1B tokens/day is a bill of about $45: a self-serve number, visible in the cost calculator before you commit a workload.
Pick a number and reach it
The tiers are published, not negotiated: the docs limits page is rendered live from the same config that answers your keys, and the 429 error message names your tier when you do hit a limit. If your verified workload needs more than 1B tokens/day, limits go up again on request: enterprise-scale traffic is now self-serve at every level this side of a signed contract.
For agents: x402
Agents do not hit signup forms. Every model is callable per-call with x402 (USDC on Base), a payment header per request, at POST /v1/agent/chat/completions; POST /v1/agent/quote returns the price of a planned call in advance. Current tier ceilings are in pricing.md and llms.txt for any machine to read.
Build bigger. Your ceiling just moved a decimal point.
James Miano
CTO & ML Engineer at Lacesse Ventures. James specializes in low-overhead LLM quantization processes, custom ternary weights architectures, and localized server optimization.
Stop paying for overpriced round-trip latency
Why route queries over Western servers when you can use low-overhead hardware located in Nairobi? Save 87% on your monthly inference spend. No minimum credit limits. Pay with card or mobile money.