Featherless AI
Flat monthly rate for unlimited use of 20,000+ open LLMs
Overview
Featherless is a serverless inference platform giving subscribers unlimited monthly access to over 20,000 open-weight Hugging Face models through one API key, billed as a flat rate rather than per token. It's built by researchers from the RWKV project and doesn't log prompts or completions.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 4 concurrent connections
- Up to 32K context
- No prompt logging
- 8 concurrent connections
- Up to 256K context
- 1 agent sandbox with persistent storage
- 8 concurrent connections
- Up to 256K context
- Larger agent sandbox
What you can produce with Featherless AI
- Unlimited monthly requests at a flat subscription rate (no per-token billing)
- Access to 6,700+ Hugging Face models, auto-onboarded at 100+ downloads
- OpenAI-compatible single API key across the full catalog
- No prompt/completion logging
- Agent runtime with persistent sandbox storage (Agent plans)
- Sub-250ms claimed model cold starts
- Per-request/credit billing option for pay-as-you-go use
ASEAN Perspective
Featherless AI in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Featherless's whole argument is that flat-rate beats metered pricing once you're making enough calls: pay $25-$200 a month and get unlimited requests against whatever's in the catalog, rather than watching a per-token bill climb. That catalog is the other selling point — Featherless auto-onboards any Hugging Face model with 100+ downloads, which is how it became HF's largest inference provider by model count (6,700+ and counting).
The trade-off shows up at the edges: concurrency is capped by plan (4 units on the $25 tier, 8 on the higher ones), so bursty or high-parallelism workloads need to buy up, and cold starts on rarely-used models are a real cost of running a long-tail catalog. There's no free tier to kick the tyres, and automated trust-scoring services disagree sharply on legitimacy — worth doing your own diligence before committing a card.
What people say
Featherless was built by researchers who'd worked on the RWKV project, and it raised a $20M Series A pitched explicitly as building "the neutral foundation for open-source AI" — a jab at the vertically-integrated labs. The company's core wager is architectural: instead of provisioning fixed GPU capacity per model, it dynamically loads and swaps models on demand, which is how it can plausibly offer tens of thousands of them without pre-committing hardware to each one.
That catalog is the headline number. Featherless says it's Hugging Face's largest LLM inference provider by model count, serving 6,700+ open-weight models with a policy that anything on HF with 100+ downloads gets auto-onboarded. Cold starts are claimed to average under 250ms, which matters a lot when you're swapping between thousands of rarely-called models rather than serving a handful of hot ones.
Pricing is flat and subscription-based rather than metered: Premium runs $25/month, Agent Standard $100/month, Agent Pro $200/month, each stepping up concurrency (4 to 8 units) and context length (32K to 256K). A separate per-request/credit option exists for teams that want to pay for what they use rather than a flat fee. As of April 2026 the company claimed roughly 10,000 customers spanning individual developers on entry plans up to enterprises paying $1-2 million a year — a wide range that suggests the flat-rate model scales further up the stack than a typical prosumer subscription.
Independent reception is mixed and thin. Automated trust-scoring services disagree with each other by a wide margin — GridinSoft rated it 91/100 legitimate, ScamAdviser put it at 66%, while Scam Detector flagged it as "New, Suspicious, and Dubious" at 18.1 — which mostly says these scoring tools are unreliable for a company this size, not that any one of them is right. On the user-review side, at least one public complaint described text generation not working and worse output than free Hugging Face access despite paying, which is worth weighing against the no-logging privacy pitch that's otherwise hard for an outsider to verify independently.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Featherless AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Featherless AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Featherless AI directly →
Spotted something out of date? Suggest an update →
Alternatives to Featherless AI
More in LLMs & Chat
