Featherless
Serverless API access to a broad open-weight model catalogue, with capacity subscriptions and per-request billing.
About Featherless
featherless is a serverless inference platform for developers and teams seeking open-weight language models without running model-serving infrastructure. A single API key connects applications to a catalogue containing tens of thousands of models, which makes it practical to test alternatives while keeping the integration stable. Paid plans differ in model-size access, concurrency, context, and billing method. The 229B Full Context and Max plans sell capacity per unit each month, while Feather Per-Request uses prepaid credits and charges for successful requests based on model and token usage. Eligible paid tiers can also launch preconfigured apps and agents in managed cloud sandboxes. The privacy policy says featherless records usage totals but does not collect or store prompts or completions. This makes the service a credible fit for model exploration and managed inference, though teams still need to match plan limits and variable per-request costs to their workload.
Best for
Best use cases
A prototype needs to switch among model families while keeping the client integration unchanged.
A production application needs up to 256K context and predictable concurrent inference units.
A team wants to launch a packaged assistant in a managed sandbox within its paid plan.
Compare before choosing
Choose the 229B plan when its model-size ceiling is sufficient and Max access is unnecessary.
Choose Max when access to any catalogue model matters more than the higher monthly unit cost.
Choose Per-Request for higher concurrency and usage-based billing, while accounting for variable token costs.
Features
Broad model catalogue
Use one API key to access tens of thousands of open-weight models without deploying each model yourself.
Serverless API
Send inference requests without provisioning or maintaining dedicated model-serving infrastructure.
Managed agent marketplace
Launch preconfigured cloud agents with Featherless inference already connected on eligible paid plans.
Pros
- One API integration gives developers access to tens of thousands of open-weight models.
- Managed serverless inference removes the cost and workflow burden of operating model-serving infrastructure.
- The privacy policy says prompts and completions are not collected or stored.
Cons
- There is no verified free plan; the listed per-request option starts at $25 in prepaid monthly credits.
- Capacity plans impose concurrency limits, while per-request costs vary with model choice and token usage.
- Agent sandboxes are limited to paid plans above the entry $25 tier, which raises the cost of that workflow.
Pricing Plans
Featherless 229B Full Context
- Private, secure and anonymous usage with no prompt or completion logs
- Models up to 229B parameters
- 8 concurrent units
- Context up to 256K
- One sandbox per unit
Featherless Max
- Private, secure and anonymous usage with no prompt or completion logs
- Access to any model in the catalogue
- 8 concurrent units
- Context up to 256K
- One sandbox per unit
Feather Per-Request
- Credit-based API access
- Charges for successful requests based on model price and token use
- No model-size limit
- 100 concurrent units
* Pricing information is for reference only. Please refer to the latest announcement on Featherless's official website.