specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Featherless AI
providerId: featherless
created: '2026-06-21'
modified: '2026-06-21'
reconciled: false
tags:
- AI
- LLM
- Inference
- Serverless
- Open Models
- Rate Limiting
- Quotas
- Throttling
description: >-
  Featherless AI does not meter tokens. Instead each subscription plan grants
  a fixed number of concurrent connections (concurrency limit), and tokens
  are unlimited within that concurrency. Smaller models allow higher effective
  throughput than larger ones. The number of in-flight requests an account
  may run is therefore the primary limit; additional requests beyond the plan
  concurrency are rejected until a slot frees. Live concurrency state can be
  observed via the account concurrency stream.
notes: >-
  Verify per-plan concurrency on the Featherless plans/concurrency pages on
  reconciliation; values change as users move between Basic, Premium, agent,
  and business tiers.
sources:
- https://featherless.ai/docs/concurrency
- https://featherless.ai/docs/concurrency-limits
- https://featherless.ai/docs/plans
responseCodes:
  throttled: 429
limits:
- name: Concurrent Connections
  scope: account
  metric: connections
  limit: see plan (Basic 2, Premium 4, agent/business 8 per unit)
  notes: Maximum simultaneous in-flight inference requests per account/plan.
- name: Tokens
  scope: account
  metric: tokens
  limit: unlimited within plan concurrency
  notes: Featherless bills a flat monthly subscription; tokens are not metered.
- name: Model Size
  scope: account
  metric: parameters
  limit: see plan (Basic up to 15B; Premium any size)
  notes: Larger models reduce effective per-connection throughput.
policies:
- name: Concurrency-Based Limiting
  description: >-
    Throughput is governed by the number of concurrent connections in the
    plan rather than by token quotas; excess concurrent requests are rejected
    with HTTP 429 until a slot is available.
- name: Backoff Strategy
  description: Clients should implement exponential backoff with jitter and honor Retry-After when concurrency is saturated.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com